diff --git a/.github/workflows/build-artifacts.yml b/.github/workflows/build-artifacts.yml
index d642a81117..dca60a9d43 100644
--- a/.github/workflows/build-artifacts.yml
+++ b/.github/workflows/build-artifacts.yml
@@ -64,7 +64,7 @@ jobs:
     strategy:
       matrix:
         os: [macos-latest, ubuntu-latest, windows-latest]
-        python-version: ["3.10", "3.11", "3.12", "3.13"]
+        python-version: ["3.10", "3.11", "3.12", "3.13", "3.14"]
     steps:
       - uses: actions/checkout@v4
       - name: Set up Python ${{ matrix.python-version }}
@@ -245,8 +245,8 @@ jobs:
       - name: Generate json schema
         run: |
           mkdir /tmp/json-schemas
-          uv run python -c "from dstack._internal.core.models.configurations import DstackConfiguration; print(DstackConfiguration.schema_json())" > /tmp/json-schemas/configuration.json
-          uv run python -c "from dstack._internal.core.models.profiles import ProfilesConfig; print(ProfilesConfig.schema_json())" > /tmp/json-schemas/profiles.json
+          uv run python -c "from dstack._internal.core.models.configurations import DstackConfiguration; import json; print(json.dumps(DstackConfiguration.model_json_schema()))" > /tmp/json-schemas/configuration.json
+          uv run python -c "from dstack._internal.core.models.profiles import ProfilesConfig; import json; print(json.dumps(ProfilesConfig.model_json_schema()))" > /tmp/json-schemas/profiles.json
       - uses: actions/upload-artifact@v4
         with:
           name: json-schemas
diff --git a/pyproject.toml b/pyproject.toml
index 14f40e082e..60b233223d 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -27,16 +27,14 @@ dependencies = [
     "rich-argparse",
     "tqdm",
     "questionary>=2.0.1",
-    "pydantic>=1.10.10,<2.0.0",
-    "pydantic-duality>=1.2.4",
+    "pydantic>=2.12",
     "websocket-client",
     "python-multipart>=0.0.16",
     "filelock",
     "psutil",
-    "gpuhunt==0.1.27",
+    "gpuhunt==0.1.29",
     "argcomplete>=3.5.0",
     "ignore-python>=0.2.0",
-    "orjson",
     "apscheduler<4",
 ]
 
@@ -115,7 +113,6 @@ include = [
     "src/dstack/_internal/core/backends/runpod",
     "src/dstack/_internal/core/backends/slurm",
     "src/dstack/_internal/cli/services/configurators",
-    "src/dstack/_internal/cli/services/endpoints",
     "src/dstack/_internal/cli/commands",
     "src/tests/_internal/server/background/pipeline_tasks",
 ]
@@ -140,9 +137,12 @@ env = [
     "DSTACK_SSHPROXY_API_TOKEN=test-token",
 ]
 filterwarnings = [
+    # Fail on any use of a pydantic v1 shim (`.dict()`, `parse_obj_as`, class-based `Config`, ...)
+    # so they cannot creep back after the v2 migration.
+    "error::pydantic.PydanticDeprecatedSince20",
     # testcontainers modules use deprecated decorators – nothing we can do:
     # https://github.com/testcontainers/testcontainers-python/issues/874
-    "ignore:^The @wait_container_is_ready decorator:DeprecationWarning"
+    "ignore:^The @wait_container_is_ready decorator:DeprecationWarning",
 ]
 
 [dependency-groups]
diff --git a/scripts/add_backend.py b/scripts/add_backend.py
index a18e48c7f2..9987d9ce9a 100644
--- a/scripts/add_backend.py
+++ b/scripts/add_backend.py
@@ -1,8 +1,15 @@
 import argparse
 from pathlib import Path
+from typing import Optional
 
 import jinja2
 
+TEMPLATE_DIR_PATH = Path(__file__).parent.parent.joinpath(
+    "src/dstack/_internal/core/backends/template"
+)
+BACKENDS_DIR_PATH = Path(__file__).parent.parent.joinpath("src/dstack/_internal/core/backends")
+TEMPLATE_FILENAMES = ["backend.py", "compute.py", "configurator.py", "models.py"]
+
 
 def main():
     parser = argparse.ArgumentParser(
@@ -21,25 +28,34 @@ def main():
     generate_backend_code(args.name)
 
 
-def generate_backend_code(backend_name: str):
-    template_dir_path = Path(__file__).parent.parent.joinpath(
-        "src/dstack/_internal/core/backends/template"
-    )
+def generate_backend_code(backend_name: str, backends_dir_path: Optional[Path] = None) -> Path:
+    """
+    Renders the scaffold templates for a new backend.
+
+    Args:
+        backend_name: The backend name in CamelCase, e.g. `VastAI`.
+        backends_dir_path: Where to create the backend package. Defaults to the real backends
+            directory; tests pass a temporary one.
+
+    Returns:
+        The path of the generated backend package.
+    """
     env = jinja2.Environment(
         loader=jinja2.FileSystemLoader(
-            searchpath=template_dir_path,
+            searchpath=TEMPLATE_DIR_PATH,
         ),
         keep_trailing_newline=True,
     )
-    backend_dir_path = Path(__file__).parent.parent.joinpath(
-        f"src/dstack/_internal/core/backends/{backend_name.lower()}"
-    )
-    backend_dir_path.mkdir(exist_ok=True)
-    for filename in ["backend.py", "compute.py", "configurator.py", "models.py"]:
+    if backends_dir_path is None:
+        backends_dir_path = BACKENDS_DIR_PATH
+    backend_dir_path = backends_dir_path.joinpath(backend_name.lower())
+    backend_dir_path.mkdir(parents=True, exist_ok=True)
+    for filename in TEMPLATE_FILENAMES:
         template = env.get_template(f"{filename}.jinja")
         with open(backend_dir_path.joinpath(filename), "w+") as f:
             f.write(template.render({"backend_name": backend_name}))
     backend_dir_path.joinpath("__init__.py").write_text("")
+    return backend_dir_path
 
 
 if __name__ == "__main__":
diff --git a/scripts/docs/gen_openapi_reference.py b/scripts/docs/gen_openapi_reference.py
index def7cf24b6..9998e1b7be 100644
--- a/scripts/docs/gen_openapi_reference.py
+++ b/scripts/docs/gen_openapi_reference.py
@@ -24,7 +24,10 @@
 TAG_LIST_END = "<!-- END GENERATED HTTP API TAGS -->"
 HTTP_METHODS = {"get", "put", "post", "delete", "options", "head", "patch", "trace"}
 UNTAGGED_TAG = "default"
-OPENAPI_VERSION = "3.0.3"
+# Must stay 3.1.x: pydantic v2 generates JSON Schema draft 2020-12, so the spec contains
+# constructs 3.0 has no equivalent for (`{"type": "null"}` for an optional field, `const`).
+# Declaring 3.0.3 over those produces a spec no validator accepts. swagger-ui renders 3.1.
+OPENAPI_VERSION = "3.1.0"
 
 if os.environ.get(disable_env):
     logger.warning("OpenAPI reference generation is disabled")
diff --git a/scripts/docs/gen_schema_reference.py b/scripts/docs/gen_schema_reference.py
index 62f379c821..1c9e9d3712 100644
--- a/scripts/docs/gen_schema_reference.py
+++ b/scripts/docs/gen_schema_reference.py
@@ -13,7 +13,8 @@
 
 import mkdocs_gen_files
 import yaml
-from pydantic.main import BaseModel
+from pydantic import BaseModel, RootModel
+from pydantic_core import PydanticUndefined
 from typing_extensions import Annotated, Any, Dict, Literal, Type, Union, get_args, get_origin
 
 from dstack._internal.core.models.resources import Range
@@ -25,20 +26,65 @@
 logger.info("Generating schema reference...")
 
 
-def _is_linkable_type(annotation: Any) -> bool:
-    """Check if a type annotation contains a BaseModel subclass (excluding Range)."""
+def _unwrap_optional(annotation: Any) -> Any:
+    """The non-`None` member of an `Optional[...]`, or the annotation unchanged."""
+    if get_origin(annotation) is Union:
+        args = [a for a in get_args(annotation) if a is not type(None)]
+        if len(args) == 1:
+            return args[0]
+    return annotation
+
+
+def _linkable_model(annotation: Any) -> Optional[type]:
+    """
+    The `BaseModel` subclass a field links to in the reference, if any.
+
+    pydantic v2 strips `Annotated` off `FieldInfo.annotation`, so the shape this used to unwrap by
+    hand (`Annotated[Optional[SSHParams], Field(...)]`) now arrives as plain `Optional[SSHParams]`.
+    Recursing over the annotation covers both, and also catches a bare model field, which the old
+    `get_args(...)[0]` approach silently missed.
+    """
     origin = get_origin(annotation)
+    # The container cases come first: `get_origin(Annotated[X, ...])` is `Annotated`, which is
+    # itself a class, so testing `inspect.isclass` first would stop before ever unwrapping it.
+    if origin in (Annotated, Union, list):
+        for arg in get_args(annotation):
+            if arg is type(None):
+                continue
+            found = _linkable_model(arg)
+            if found is not None:
+                return found
+        return None
     type_ = origin if origin is not None else annotation
-    if inspect.isclass(type_):
-        return issubclass(type_, BaseModel) and not issubclass(type_, Range)
-    if origin is Annotated:
-        return _is_linkable_type(get_args(annotation)[0])
-    if origin is Union:
-        return any(_is_linkable_type(arg) for arg in get_args(annotation))
-    if origin is list:
-        args = get_args(annotation)
-        return bool(args) and _is_linkable_type(args[0])
-    return False
+    if inspect.isclass(type_) and issubclass(type_, BaseModel) and not issubclass(type_, Range):
+        return type_
+    return None
+
+
+# Scalar JSON Schema types, mapped to how the docs spell them. Deliberately an allowlist rather
+# than a full mapping: `array` and `object` would only restate what the annotation already renders
+# more precisely (`list[str]` gaining a bare `list`, `dict` gaining `object`), and merging anything
+# into a bracketed type corrupts it, since `_enrich_type_from_schema` splits the rendered type on
+# `" | "` — which `list["no-capacity" | "interruption"]` contains.
+_ENRICHABLE = {"string": "str", "integer": "int", "boolean": "bool", "number": "float"}
+
+
+def _shorthand_primitives(model: Type) -> list:
+    """
+    The primitive types a model accepts in place of its object form, e.g. `8` or `arm:8` for
+    `CPUSpec`. Taken from the model's own validation JSON Schema, which is the same declaration
+    that produces the published `configuration.json`.
+    """
+    try:
+        schema = model.model_json_schema(mode="validation")
+    except Exception:
+        return []
+    found = {
+        _ENRICHABLE[entry["type"]]
+        for entry in schema.get("anyOf", [])
+        if entry.get("type") in _ENRICHABLE
+    }
+    return sorted(found, key=_type_sort_key)
 
 
 def _type_sort_key(t: str) -> tuple:
@@ -57,7 +103,7 @@ def _type_sort_key(t: str) -> tuple:
     return (5, t)
 
 
-def get_friendly_type(annotation: Type) -> str:
+def get_friendly_type(annotation: Any) -> str:
     """Get a user-friendly type string for documentation.
 
     Produces types like: ``int | str``, ``"vscode" | "cursor"``, ``list[object]``.
@@ -112,10 +158,11 @@ def get_friendly_type(annotation: Type) -> str:
 
         # Range — depends on inner type parameter
         if issubclass(annotation, Range):
-            min_field = annotation.__fields__.get("min")
-            if min_field and inspect.isclass(min_field.type_):
+            min_field = annotation.model_fields.get("min")
+            inner = _unwrap_optional(min_field.annotation) if min_field else None
+            if inspect.isclass(inner):
                 # Range[Memory] → str, Range[int] → int | str
-                if issubclass(min_field.type_, float):
+                if issubclass(inner, float):
                     return "str"
             return "int | str"
 
@@ -127,13 +174,16 @@ def get_friendly_type(annotation: Type) -> str:
 
         # BaseModel subclass (not Range)
         if issubclass(annotation, BaseModel) and not issubclass(annotation, Range):
-            # Root models (with __root__ field) — resolve from the root type
-            if "__root__" in annotation.__fields__:
-                return get_friendly_type(annotation.__fields__["__root__"].annotation)
-            # Models with custom __get_validators__ accept primitive input (int, str)
-            # in addition to the full object form (e.g., GPUSpec, CPUSpec, DiskSpec)
-            if "__get_validators__" in annotation.__dict__:
-                return "int | str | object"
+            # Root models — resolve from the root type
+            if issubclass(annotation, RootModel):
+                return get_friendly_type(annotation.model_fields["root"].annotation)
+            # Models that define their own core schema also accept a shorthand. Read which
+            # primitives from the model's own JSON Schema rather than assuming `int | str`:
+            # `CPUSpec` takes both, but `FilePathMapping` and `RepoSpec` take only a string.
+            if "__get_pydantic_core_schema__" in annotation.__dict__:
+                shorthand = _shorthand_primitives(annotation)
+                if shorthand:
+                    return " | ".join([*shorthand, "object"])
             return "object"
 
         # ComputeCapability (tuple subclass that parses "7.5" strings)
@@ -163,33 +213,24 @@ def get_friendly_type(annotation: Type) -> str:
     return str(annotation)
 
 
-_JSON_SCHEMA_TYPE_MAP = {
-    "string": "str",
-    "integer": "int",
-    "number": "float",
-    "boolean": "bool",
-    "array": "list",
-    "object": "object",
-}
-
-
 def _enrich_type_from_schema(friendly_type: str, prop_schema: Dict[str, Any]) -> str:
     """Enrich the friendly type with extra accepted types from the JSON schema.
 
-    Models may define ``schema_extra`` that adds ``anyOf`` entries for fields
-    that accept alternative input types (e.g., duration fields typed as ``int``
-    but also accepting ``str`` like ``"5m"``).
+    A field's annotation is its *post-validation* type, so it does not show what a before-validator
+    also accepts — a duration typed ``int`` takes ``"5m"``, ``false`` and ``"off"`` as well. Those
+    come from the type's ``json_schema_input_type``, i.e. the same declaration that produces the
+    published schema.
     """
     any_of = prop_schema.get("anyOf")
     if not any_of:
         return friendly_type
-    # Only consider string/integer — the most common alternative input types.
-    # Skip boolean (typically a backward-compat artifact) and object/array.
-    _ENRICHABLE = {"string": "str", "integer": "int"}
     schema_types = set()
     for entry in any_of:
-        # Skip entries with enum constraints — those are already captured as literal values
-        if "enum" in entry:
+        # A single accepted value (`Literal["off"]`) is more useful spelled out than as `str`.
+        # Duplicates are removed below, so an annotation that already shows it is unaffected.
+        literals = [entry["const"]] if "const" in entry else entry.get("enum", [])
+        if literals:
+            schema_types.update(f'"{v}"' for v in literals if isinstance(v, str))
             continue
         mapped = _ENRICHABLE.get(entry.get("type", ""))
         if mapped:
@@ -200,9 +241,6 @@ def _enrich_type_from_schema(friendly_type: str, prop_schema: Dict[str, Any]) ->
     if not new_parts:
         return friendly_type
     all_parts = list(set(current_parts) | new_parts)
-    # If str is now present, single-value literals are redundant
-    if "str" in all_parts:
-        all_parts = [p for p in all_parts if not p.startswith('"') or p in all_parts]
     all_parts.sort(key=_type_sort_key)
     return " | ".join(all_parts)
 
@@ -228,15 +266,17 @@ def generate_schema_reference(
                 "",
             ]
         )
-    # Get JSON schema to detect extra accepted types from schema_extra
+    # The schema says what a field *accepts*, which is wider than its annotation wherever a
+    # before-validator coerces. `mode="validation"` is pydantic's default, but state it: the
+    # serialization schema carries the narrow type and would defeat the whole point.
     try:
-        schema_props = cls.schema().get("properties", {})
+        schema_props = cls.model_json_schema(mode="validation").get("properties", {})
     except Exception:
         schema_props = {}
-    for name, field in cls.__fields__.items():
+    for name, field in cls.model_fields.items():
         default = field.default
         default_repr: Optional[str]
-        if default is None:
+        if default is None or default is PydanticUndefined:
             default_repr = None
         elif isinstance(default, (list, tuple, dict)) and len(default) == 0:
             default_repr = None
@@ -252,24 +292,17 @@ def generate_schema_reference(
         friendly_type = _enrich_type_from_schema(friendly_type, schema_props.get(name, {}))
         values = dict(
             name=name,
-            description=field.field_info.description,
+            description=field.description,
             type=friendly_type,
             default=default_repr,
-            required=field.required,
+            required=field.is_required(),
         )
         # TODO: If the field doesn't have description (e.g. BaseConfiguration.type), we could fallback to docstring
         if values["description"]:
             if overrides and name in overrides:
                 values.update(overrides[name])
-            field_type = next(iter(get_args(field.annotation)), None)
-            # TODO: This is a dirty workaround
-            if field_type:
-                if field.annotation.__name__ == "Annotated":
-                    if field_type.__name__ in ["Optional", "List", "list", "Union"]:
-                        field_type = get_args(field_type)[0]
-                base_model = _is_linkable_type(field_type)
-            else:
-                base_model = False
+            field_type = _linkable_model(field.annotation)
+            base_model = field_type is not None
             _defaults = (
                 f"Defaults to `{values['default']}`."
                 if not base_model and values.get("default")
diff --git a/src/dstack/__init__.py b/src/dstack/__init__.py
index a90955ea78..e69de29bb2 100644
--- a/src/dstack/__init__.py
+++ b/src/dstack/__init__.py
@@ -1,4 +0,0 @@
-import sys
-
-if sys.version_info >= (3, 14):
-    raise ImportError("dstack does not support Python 3.14 or later. Please use Python 3.10–3.13.")
diff --git a/src/dstack/_internal/cli/commands/fleet.py b/src/dstack/_internal/cli/commands/fleet.py
index c0b4c0e715..8e723cb41b 100644
--- a/src/dstack/_internal/cli/commands/fleet.py
+++ b/src/dstack/_internal/cli/commands/fleet.py
@@ -15,7 +15,6 @@
 from dstack._internal.cli.utils.fleet import get_fleets_table, print_fleets_table
 from dstack._internal.core.errors import CLIError, ResourceNotExistsError
 from dstack._internal.core.models.common import EntityReference
-from dstack._internal.utils.json_utils import pydantic_orjson_dumps_with_indent
 
 
 class FleetCommand(APIBaseCommand):
@@ -176,4 +175,4 @@ def _get(self, args: argparse.Namespace):
             console.print(f"Fleet [code]{args.name or args.id}[/] not found")
             exit(1)
 
-        print(pydantic_orjson_dumps_with_indent(fleet.dict(), default=None))
+        print(fleet.model_dump_json(indent=2))
diff --git a/src/dstack/_internal/cli/commands/gateway.py b/src/dstack/_internal/cli/commands/gateway.py
index 71661f3f00..2489e726b4 100644
--- a/src/dstack/_internal/cli/commands/gateway.py
+++ b/src/dstack/_internal/cli/commands/gateway.py
@@ -20,7 +20,6 @@
 from dstack._internal.core.errors import CLIError
 from dstack._internal.core.models.common import EntityReference
 from dstack._internal.core.models.gateways import GatewayStatus
-from dstack._internal.utils.json_utils import pydantic_orjson_dumps_with_indent
 from dstack._internal.utils.logging import get_logger
 
 logger = get_logger(__name__)
@@ -206,4 +205,4 @@ def _get(self, args: argparse.Namespace):
             gateway_project=args.name.project or self.api.project,
             gateway_name=args.name.name,
         )
-        print(pydantic_orjson_dumps_with_indent(gateway.dict(), default=None))
+        print(gateway.model_dump_json(indent=2))
diff --git a/src/dstack/_internal/cli/commands/offer.py b/src/dstack/_internal/cli/commands/offer.py
index 56c7cd2b56..5d697c5393 100644
--- a/src/dstack/_internal/cli/commands/offer.py
+++ b/src/dstack/_internal/cli/commands/offer.py
@@ -252,4 +252,4 @@ def _print_offers_json(run_plan: RunPlan):
         offers=job_plan.offers,
         total_offers=job_plan.total_offers,
     )
-    print(output.json())
+    print(output.model_dump_json())
diff --git a/src/dstack/_internal/cli/commands/preset.py b/src/dstack/_internal/cli/commands/preset.py
index 184df2592b..ee0e2690bd 100644
--- a/src/dstack/_internal/cli/commands/preset.py
+++ b/src/dstack/_internal/cli/commands/preset.py
@@ -230,7 +230,7 @@ def _list(self, args: argparse.Namespace) -> None:
         if args.json:
             self._reconcile()
             presets = _filter_presets(PresetStore().list(), base=base, repo=repo)
-            print(PresetListOutput(presets=presets).json())
+            print(PresetListOutput(presets=presets).model_dump_json())
             return
         verbose = args.verbose
         if not getattr(args, "watch", False):
@@ -330,7 +330,7 @@ def _get(self, args: argparse.Namespace) -> None:
         preset = PresetStore().find_by_id_or_name(args.preset)
         if preset is None:
             raise CLIError(f"Preset {args.preset!r} does not exist")
-        print(preset.json())
+        print(preset.model_dump_json())
 
     def _apply(self, args: argparse.Namespace) -> None:
         self._reconcile()
@@ -493,8 +493,8 @@ def _get_effective_configuration(
     if getattr(args, "max_trials", None) is not None:
         configuration.max_trials = args.max_trials
     profile = load_profile_from_args(args=args, repo_dir=Path.cwd())
-    for field in ProfileParams.__fields__:
+    for field in ProfileParams.model_fields:
         if getattr(configuration, field) is None:
             setattr(configuration, field, getattr(profile, field))
     apply_profile_args(args, configuration)
-    return PresetConfiguration.parse_obj(configuration.dict())
+    return PresetConfiguration.model_validate(configuration.model_dump())
diff --git a/src/dstack/_internal/cli/commands/run.py b/src/dstack/_internal/cli/commands/run.py
index 337b0a75cf..909d2dbb27 100644
--- a/src/dstack/_internal/cli/commands/run.py
+++ b/src/dstack/_internal/cli/commands/run.py
@@ -5,7 +5,6 @@
 from dstack._internal.cli.services.completion import RunNameCompleter
 from dstack._internal.cli.utils.common import console
 from dstack._internal.core.errors import CLIError, ResourceNotExistsError
-from dstack._internal.utils.json_utils import pydantic_orjson_dumps_with_indent
 
 
 class RunCommand(APIBaseCommand):
@@ -66,4 +65,4 @@ def _get(self, args: argparse.Namespace):
             console.print(f"Run [code]{args.name or args.id}[/] not found")
             exit(1)
 
-        print(pydantic_orjson_dumps_with_indent(run.dict(), default=None))
+        print(run.model_dump_json(indent=2))
diff --git a/src/dstack/_internal/cli/commands/volume.py b/src/dstack/_internal/cli/commands/volume.py
index e78ec352c6..71c51559cc 100644
--- a/src/dstack/_internal/cli/commands/volume.py
+++ b/src/dstack/_internal/cli/commands/volume.py
@@ -13,7 +13,6 @@
 )
 from dstack._internal.cli.utils.volume import get_volumes_table, print_volumes_table
 from dstack._internal.core.errors import ResourceNotExistsError
-from dstack._internal.utils.json_utils import pydantic_orjson_dumps_with_indent
 
 
 class VolumeCommand(APIBaseCommand):
@@ -114,4 +113,4 @@ def _get(self, args: argparse.Namespace):
             console.print("Volume not found")
             exit(1)
 
-        print(pydantic_orjson_dumps_with_indent(volume.dict(), default=None))
+        print(volume.model_dump_json(indent=2))
diff --git a/src/dstack/_internal/cli/models/configurations.py b/src/dstack/_internal/cli/models/configurations.py
index b0e08942b7..f7a3554246 100644
--- a/src/dstack/_internal/cli/models/configurations.py
+++ b/src/dstack/_internal/cli/models/configurations.py
@@ -1,15 +1,18 @@
 from typing import Annotated, Any, Literal, Optional, Union
 
-from pydantic import Field, PositiveInt, root_validator, validator
+from pydantic import (
+    Field,
+    PositiveInt,
+    field_validator,
+    model_validator,
+)
 
 from dstack._internal.core.models.common import (
     CoreModel,
     EntityReference,
-    generate_dual_core_model,
 )
 from dstack._internal.core.models.envs import Env
-from dstack._internal.core.models.profiles import ProfileParams, ProfileParamsConfig
-from dstack._internal.utils.json_schema import add_extra_schema_types
+from dstack._internal.core.models.profiles import ProfileParams
 
 DEFAULT_CONCURRENCY = 8
 
@@ -32,11 +35,13 @@ def exact_repo(self) -> str:
     def allows_variant_selection(self) -> bool:
         return False
 
-    @validator("repo")
+    @field_validator("repo")
+    @classmethod
     def validate_repo(cls, value: str) -> str:
         return _validate_model(value, field="repo")
 
-    @validator("name")
+    @field_validator("name")
+    @classmethod
     def validate_name(cls, value: Optional[str]) -> Optional[str]:
         if value is None:
             return None
@@ -61,7 +66,8 @@ def exact_repo(self) -> None:
     def allows_variant_selection(self) -> bool:
         return True
 
-    @validator("base")
+    @field_validator("base")
+    @classmethod
     def validate_base(cls, value: str) -> str:
         return _validate_model(value, field="base")
 
@@ -77,26 +83,16 @@ class PresetPromptFile(CoreModel):
         Field(description="The path to a prompt file, relative to the configuration file"),
     ]
 
-    @validator("path")
+    @field_validator("path")
+    @classmethod
     def validate_path(cls, value: str) -> str:
         if not value.strip():
             raise ValueError("Prompt path must be a non-empty string")
         return value
 
 
-class PresetConfigurationConfig(ProfileParamsConfig):
-    @staticmethod
-    def schema_extra(schema: dict[str, Any]):
-        ProfileParamsConfig.schema_extra(schema)
-        add_extra_schema_types(
-            schema["properties"]["model"],
-            extra_types=[{"type": "string"}],
-        )
-
-
 class PresetConfiguration(
     ProfileParams,
-    generate_dual_core_model(PresetConfigurationConfig),
 ):
     type: Annotated[Literal["preset"], Field(description="The configuration type")] = "preset"
     name: Annotated[
@@ -158,11 +154,12 @@ class PresetConfiguration(
     gateway: Annotated[
         Optional[Union[bool, EntityReference, str]],
         Field(
+            union_mode="left_to_right",  # preserving pydantic v1 parsing behavior
             description=(
                 "The name of the gateway. Specify boolean `false` to run without a gateway."
                 " Specify boolean `true` to run with the default gateway."
                 " Omit to run with the default gateway if there is one, or without a gateway otherwise"
-            )
+            ),
         ),
     ] = None
     env: Annotated[Env, Field(description="The mapping or the list of environment variables")] = (
@@ -173,7 +170,8 @@ class PresetConfiguration(
     def effective_concurrency(self) -> int:
         return self.concurrency if self.concurrency is not None else DEFAULT_CONCURRENCY
 
-    @root_validator(pre=True)
+    @model_validator(mode="before")
+    @classmethod
     def apply_model_shorthand(cls, values: Any) -> Any:
         if not isinstance(values, dict):
             return values
@@ -189,13 +187,15 @@ def apply_model_shorthand(cls, values: Any) -> Any:
             values["model"] = {"base": base} if base else {"repo": repo}
         return values
 
-    @validator("model", pre=True)
+    @field_validator("model", mode="before", json_schema_input_type=Union[PresetModelSpec, str])
+    @classmethod
     def parse_model(cls, value: Any) -> Any:
         if isinstance(value, str):
             return {"repo": _validate_model(value, field="model")}
         return value
 
-    @validator("prompt")
+    @field_validator("prompt")
+    @classmethod
     def validate_prompt(cls, value: Any) -> Any:
         if isinstance(value, str):
             if not value.strip():
@@ -214,7 +214,7 @@ class PresetConstraints(CoreModel):
     context_length: Optional[PositiveInt] = None
     max_trials: PositiveInt
     concurrency: PositiveInt
-    fleets: list[str] = Field(min_items=1)
+    fleets: list[str] = Field(min_length=1)
     env: list[str] = []
 
 
diff --git a/src/dstack/_internal/cli/models/gateways.py b/src/dstack/_internal/cli/models/gateways.py
index 94dfa88982..1428757b98 100644
--- a/src/dstack/_internal/cli/models/gateways.py
+++ b/src/dstack/_internal/cli/models/gateways.py
@@ -1,15 +1,10 @@
 from typing import List
 
-from dstack._internal.core.models.common import CoreConfig, generate_dual_core_model
+from dstack._internal.core.models.common import CoreModel
 from dstack._internal.core.models.gateways import Gateway
-from dstack._internal.utils.json_utils import pydantic_orjson_dumps_with_indent
 
 
-class GatewayCommandOutputConfig(CoreConfig):
-    json_dumps = pydantic_orjson_dumps_with_indent
-
-
-class GatewayCommandOutput(generate_dual_core_model(GatewayCommandOutputConfig)):
+class GatewayCommandOutput(CoreModel):
     """JSON output model for `dstack gateway` command."""
 
     project: str
diff --git a/src/dstack/_internal/cli/models/offers.py b/src/dstack/_internal/cli/models/offers.py
index 56d5e21ea7..96df7a0ea6 100644
--- a/src/dstack/_internal/cli/models/offers.py
+++ b/src/dstack/_internal/cli/models/offers.py
@@ -1,17 +1,12 @@
 from typing import List, Literal, Optional
 
-from dstack._internal.core.models.common import CoreConfig, generate_dual_core_model
+from dstack._internal.core.models.common import CoreModel
 from dstack._internal.core.models.gpus import GpuGroup
 from dstack._internal.core.models.instances import InstanceOfferWithAvailability
 from dstack._internal.core.models.resources import ResourcesSpec
-from dstack._internal.utils.json_utils import pydantic_orjson_dumps_with_indent
 
 
-class OfferRequirementsConfig(CoreConfig):
-    json_dumps = pydantic_orjson_dumps_with_indent
-
-
-class OfferRequirements(generate_dual_core_model(OfferRequirementsConfig)):
+class OfferRequirements(CoreModel):
     """Profile/requirements output model for CLI commands."""
 
     resources: ResourcesSpec
@@ -20,11 +15,7 @@ class OfferRequirements(generate_dual_core_model(OfferRequirementsConfig)):
     reservation: Optional[str] = None
 
 
-class OfferCommandOutputConfig(CoreConfig):
-    json_dumps = pydantic_orjson_dumps_with_indent
-
-
-class OfferCommandOutput(generate_dual_core_model(OfferCommandOutputConfig)):
+class OfferCommandOutput(CoreModel):
     """JSON output model for `dstack offer` command."""
 
     project: str
@@ -34,11 +25,7 @@ class OfferCommandOutput(generate_dual_core_model(OfferCommandOutputConfig)):
     total_offers: int
 
 
-class OfferCommandGroupByGpuOutputConfig(CoreConfig):
-    json_dumps = pydantic_orjson_dumps_with_indent
-
-
-class OfferCommandGroupByGpuOutput(generate_dual_core_model(OfferCommandGroupByGpuOutputConfig)):
+class OfferCommandGroupByGpuOutput(CoreModel):
     """JSON output model for `dstack offer` command with GPU grouping."""
 
     project: str
diff --git a/src/dstack/_internal/cli/models/preset_agent.py b/src/dstack/_internal/cli/models/preset_agent.py
index e756941249..8c1710d76c 100644
--- a/src/dstack/_internal/cli/models/preset_agent.py
+++ b/src/dstack/_internal/cli/models/preset_agent.py
@@ -1,7 +1,8 @@
 import uuid
 from typing import Any, Dict, Optional
 
-from pydantic import PositiveInt, root_validator
+from pydantic import PositiveInt, model_validator
+from typing_extensions import Self
 
 from dstack._internal.cli.models.presets import PresetBenchmark
 from dstack._internal.core.models.common import CoreModel
@@ -100,9 +101,9 @@ class AgentFinalReport(CoreModel):
     benchmark: Optional[PresetBenchmark] = None
     failure_summary: Optional[str] = None
 
-    @root_validator
-    def validate_report(cls, values: dict) -> dict:
-        if values.get("success"):
+    @model_validator(mode="after")
+    def validate_report(self) -> Self:
+        if self.success:
             required = (
                 "run_id",
                 "run_name",
@@ -112,12 +113,12 @@ def validate_report(cls, values: dict) -> dict:
                 "context_length",
                 "benchmark",
             )
-            missing = [field for field in required if values.get(field) in (None, "")]
+            missing = [field for field in required if getattr(self, field) in (None, "")]
             if missing:
                 raise ValueError("successful agent report must include " + ", ".join(missing))
-        elif not values.get("failure_summary"):
+        elif not self.failure_summary:
             raise ValueError("failed agent report must include failure_summary")
-        return values
+        return self
 
 
 class PresetAgentInfo(CoreModel):
diff --git a/src/dstack/_internal/cli/models/presets.py b/src/dstack/_internal/cli/models/presets.py
index 16866ecc38..500bcf42b3 100644
--- a/src/dstack/_internal/cli/models/presets.py
+++ b/src/dstack/_internal/cli/models/presets.py
@@ -6,10 +6,10 @@
     Field,
     PositiveFloat,
     PositiveInt,
-    parse_obj_as,
-    root_validator,
-    validator,
+    field_validator,
+    model_validator,
 )
+from typing_extensions import Self
 
 from dstack._internal.core.models.common import CoreModel
 from dstack._internal.core.models.configurations import ServiceConfiguration
@@ -58,13 +58,15 @@ class PresetBenchmark(CoreModel):
     target: Optional[PresetBenchmarkTarget] = None
     client: Optional[PresetBenchmarkClient] = None
 
-    @validator("tool", "tool_version", "command")
+    @field_validator("tool", "tool_version", "command")
+    @classmethod
     def validate_non_empty(cls, value: str) -> str:
         if not value.strip():
             raise ValueError("value must be non-empty")
         return value
 
-    @validator("command")
+    @field_validator("command")
+    @classmethod
     def validate_command_has_no_bearer_token(cls, value: str) -> str:
         for match in re.finditer(r"(?i)\bbearer\s+([^\s\"']+)", value):
             token = match.group(1)
@@ -77,16 +79,16 @@ def validate_command_has_no_bearer_token(cls, value: str) -> str:
             raise ValueError("command must not contain a bearer token value")
         return value
 
-    @root_validator(skip_on_failure=True)
-    def validate_metrics(cls, values: dict) -> dict:
-        metrics = values.get("metrics")
-        workload = values.get("workload")
+    @model_validator(mode="after")
+    def validate_metrics(self) -> Self:
+        metrics = self.metrics
+        workload = self.workload
         assert metrics is not None and workload is not None
         if metrics.failed_requests != 0:
             raise ValueError("benchmark must not include failed requests")
         if metrics.successful_requests != workload.num_requests:
             raise ValueError("benchmark request count must match workload.num_requests")
-        return values
+        return self
 
 
 class PresetValidationReplica(CoreModel):
@@ -114,25 +116,26 @@ class Preset(CoreModel):
     service: ServiceConfiguration
     validations: list[PresetValidation]
 
-    @validator("base", "id", "model")
+    @field_validator("base", "id", "model")
+    @classmethod
     def validate_non_empty(cls, value: str) -> str:
         if not value.strip():
             raise ValueError("value must be non-empty")
         return value
 
-    @root_validator
-    def validate_preset(cls, values: dict) -> dict:
-        service = values.get("service")
-        validations = values.get("validations")
+    @model_validator(mode="after")
+    def validate_preset(self) -> Self:
+        service = self.service
+        validations = self.validations
         if service is None or validations is None:
-            return values
+            return self
         if service.model is None:
             raise ValueError("preset service must specify model")
         if any(group.resources is None for group in service.replica_groups):
             raise ValueError("preset service must specify resources")
         if service.name is not None or service.gateway is not None:
             raise ValueError("preset service must not specify name or gateway")
-        if any(getattr(service, field) is not None for field in ProfileParams.__fields__):
+        if any(getattr(service, field) is not None for field in ProfileParams.model_fields):
             raise ValueError("preset service must not specify placement constraints")
         if not validations:
             raise ValueError("preset must include validation evidence")
@@ -148,7 +151,7 @@ def validate_preset(cls, values: dict) -> dict:
                     raise ValueError("preset validation replicas must specify resources")
                 for resources in replica_group.resources:
                     _validate_exact_resources(resources)
-        return values
+        return self
 
 
 class PresetListOutput(CoreModel):
@@ -156,7 +159,7 @@ class PresetListOutput(CoreModel):
 
 
 def _validate_exact_resources(resources: ResourcesSpec) -> None:
-    cpu = parse_obj_as(CPUSpec, resources.cpu)
+    cpu = CPUSpec.model_validate(resources.cpu)
     if not _is_exact(cpu.count) or not _is_exact(resources.memory):
         raise ValueError("preset validation resources must be exact")
     if resources.disk is None or not _is_exact(resources.disk.size):
diff --git a/src/dstack/_internal/cli/models/runs.py b/src/dstack/_internal/cli/models/runs.py
index db951b752d..f467bc7a36 100644
--- a/src/dstack/_internal/cli/models/runs.py
+++ b/src/dstack/_internal/cli/models/runs.py
@@ -1,15 +1,10 @@
 from typing import List
 
-from dstack._internal.core.models.common import CoreConfig, generate_dual_core_model
+from dstack._internal.core.models.common import CoreModel
 from dstack._internal.core.models.runs import Run
-from dstack._internal.utils.json_utils import pydantic_orjson_dumps_with_indent
 
 
-class PsCommandOutputConfig(CoreConfig):
-    json_dumps = pydantic_orjson_dumps_with_indent
-
-
-class PsCommandOutput(generate_dual_core_model(PsCommandOutputConfig)):
+class PsCommandOutput(CoreModel):
     """JSON output model for `dstack ps` command."""
 
     project: str
diff --git a/src/dstack/_internal/cli/services/args.py b/src/dstack/_internal/cli/services/args.py
index a189984500..aa6f9e5f2b 100644
--- a/src/dstack/_internal/cli/services/args.py
+++ b/src/dstack/_internal/cli/services/args.py
@@ -1,7 +1,5 @@
 from typing import Dict
 
-from pydantic import parse_obj_as
-
 from dstack._internal.core.models import resources as resources
 from dstack._internal.core.models.configurations import PortMapping
 from dstack._internal.core.models.envs import EnvVarTuple
@@ -24,8 +22,8 @@ def cpu_spec(v: str) -> dict:
 
 
 def memory_spec(v: str) -> resources.Range[resources.Memory]:
-    return parse_obj_as(resources.Range[resources.Memory], v)
+    return resources.Range[resources.Memory].model_validate(v)
 
 
 def disk_spec(v: str) -> resources.DiskSpec:
-    return parse_obj_as(resources.DiskSpec, v)
+    return resources.DiskSpec.model_validate(v)
diff --git a/src/dstack/_internal/cli/services/configurators/run.py b/src/dstack/_internal/cli/services/configurators/run.py
index 41e01fb9db..df2d0b35d4 100644
--- a/src/dstack/_internal/cli/services/configurators/run.py
+++ b/src/dstack/_internal/cli/services/configurators/run.py
@@ -10,7 +10,6 @@
 from typing import Dict, List, Optional, Set, TypeVar
 
 import gpuhunt
-from pydantic import parse_obj_as
 
 from dstack._internal.cli.services.args import port_mapping
 from dstack._internal.cli.services.configurators.base import (
@@ -535,7 +534,7 @@ def validate_cpu_arch_and_image(self, conf: RunConfigurationT) -> None:
         Infers `resources.cpu.arch` if not set, requires `image` if the architecture is ARM.
         """
         # TODO: Remove in 0.20. Use conf.resources.cpu directly
-        cpu_spec = parse_obj_as(CPUSpec, conf.resources.cpu)
+        cpu_spec = CPUSpec.model_validate(conf.resources.cpu)
         arch = cpu_spec.arch
         if arch is None:
             gpu_spec = conf.resources.gpu
diff --git a/src/dstack/_internal/cli/services/presets/apply.py b/src/dstack/_internal/cli/services/presets/apply.py
index 40495a3574..fe6eb837a8 100644
--- a/src/dstack/_internal/cli/services/presets/apply.py
+++ b/src/dstack/_internal/cli/services/presets/apply.py
@@ -75,11 +75,11 @@ def _build_service(
     configuration: PresetConfiguration,
     preset: Preset,
 ) -> ServiceConfiguration:
-    service = preset.service.copy(deep=True)
+    service = preset.service.model_copy(deep=True)
     service.name = configuration.name
     service.gateway = configuration.gateway
     service.env.update(configuration.env)
-    for field in ProfileParams.__fields__:
+    for field in ProfileParams.model_fields:
         value = getattr(configuration, field)
         if value is not None:
             setattr(service, field, value)
diff --git a/src/dstack/_internal/cli/services/presets/create.py b/src/dstack/_internal/cli/services/presets/create.py
index cf14cfebf6..6c84c0c3cb 100644
--- a/src/dstack/_internal/cli/services/presets/create.py
+++ b/src/dstack/_internal/cli/services/presets/create.py
@@ -168,7 +168,7 @@ def _load_session_configuration(agent_session: PresetAgentSession) -> PresetConf
     # The session copy is canonical output, not user input: parse it without
     # the user-facing deprecation warnings.
     try:
-        return PresetConfiguration.parse_obj(
+        return PresetConfiguration.model_validate(
             yaml.safe_load(configuration_path.read_text(encoding="utf-8"))
         )
     except (OSError, ValueError) as e:
@@ -346,7 +346,7 @@ def _resolve_preset_env(
     """Resolves `EnvSentinel` entries from the process environment. Non-strict
     drops unresolvable entries instead of raising — for attach, where env values
     only feed redaction and the agent already runs."""
-    configuration = configuration.copy(deep=True)
+    configuration = configuration.model_copy(deep=True)
     resolved: dict[str, str] = {}
     for key, value in configuration.env.items():
         if isinstance(value, EnvSentinel):
@@ -357,7 +357,7 @@ def _resolve_preset_env(
                     raise ConfigurationError(str(e)) from e
         else:
             resolved[key] = value
-    configuration.env = Env.parse_obj(resolved)
+    configuration.env = Env.model_validate(resolved)
     return configuration
 
 
@@ -845,10 +845,10 @@ def _build_constraints(
     build_name: str,
     allowed_fleets: Sequence[str],
 ) -> str:
-    constraints = PresetConstraints.parse_obj(
+    constraints = PresetConstraints.model_validate(
         {
             "run_name_prefix": build_name,
-            "model": json.loads(configuration.model.json(exclude_none=True)),
+            "model": json.loads(configuration.model.model_dump_json(exclude_none=True)),
             "context_length": configuration.context_length,
             "max_trials": configuration.max_trials,
             "concurrency": configuration.effective_concurrency,
@@ -857,7 +857,7 @@ def _build_constraints(
         }
     )
     # All fields are always present; unset optional constraints render as null.
-    return json.dumps(json.loads(constraints.json()), indent=2) + "\n"
+    return json.dumps(json.loads(constraints.model_dump_json()), indent=2) + "\n"
 
 
 def _save_final_report_copy(
diff --git a/src/dstack/_internal/cli/services/presets/output.py b/src/dstack/_internal/cli/services/presets/output.py
index 4a9da5252b..35da0994fc 100644
--- a/src/dstack/_internal/cli/services/presets/output.py
+++ b/src/dstack/_internal/cli/services/presets/output.py
@@ -108,7 +108,9 @@ def _add_session(table: Table, session: dict[str, Any]) -> None:
     created_at = session.get("created_at")
     if isinstance(created_at, str):
         try:
-            created = pretty_date(datetime.fromisoformat(created_at))
+            # `Z` is what pydantic v2 emits for UTC, and `datetime.fromisoformat` only
+            # accepts it from Python 3.11; dstack supports 3.10.
+            created = pretty_date(datetime.fromisoformat(created_at.replace("Z", "+00:00")))
         except ValueError:
             created = created_at
     benchmark = ""
diff --git a/src/dstack/_internal/cli/services/presets/presets.py b/src/dstack/_internal/cli/services/presets/presets.py
index 3b4db39ee6..983c8ba821 100644
--- a/src/dstack/_internal/cli/services/presets/presets.py
+++ b/src/dstack/_internal/cli/services/presets/presets.py
@@ -29,10 +29,10 @@ def build_preset(
     preset_id: Optional[str] = None,
     name: Optional[str] = None,
 ) -> Preset:
-    service = service.copy(deep=True)
+    service = service.model_copy(deep=True)
     service.name = None
     service.gateway = None
-    for field in ProfileParams.__fields__:
+    for field in ProfileParams.model_fields:
         setattr(service, field, None)
     validation = PresetValidation(
         replicas=validation_replicas,
@@ -76,7 +76,8 @@ def preset_to_data(preset: Preset) -> dict[str, Any]:
         "created_at": preset.created_at.isoformat(),
         "service": service_configuration_to_preset_data(preset.service),
         "validations": [
-            json.loads(validation.json(exclude_none=True)) for validation in preset.validations
+            json.loads(validation.model_dump_json(exclude_none=True))
+            for validation in preset.validations
         ],
     }
 
@@ -84,11 +85,11 @@ def preset_to_data(preset: Preset) -> dict[str, Any]:
 def service_configuration_to_preset_data(
     configuration: ServiceConfiguration,
 ) -> dict[str, Any]:
-    service_data = json.loads(configuration.json(exclude_none=True))
+    service_data = json.loads(configuration.model_dump_json(exclude_none=True))
     service_data.pop("type", None)
     service_data.pop("name", None)
     service_data.pop("gateway", None)
-    for field in ProfileParams.__fields__:
+    for field in ProfileParams.model_fields:
         service_data.pop(field, None)
     if configuration.env:
         service_data["env"] = [
@@ -129,7 +130,7 @@ def resources_spec_from_instance_resources(resources: Resources) -> ResourcesSpe
             data["gpu"]["vendor"] = first_gpu.vendor.value
     else:
         data["gpu"] = 0
-    return ResourcesSpec.parse_obj(data)
+    return ResourcesSpec.model_validate(data)
 
 
 def set_service_gpu_vendors_from_validations(
diff --git a/src/dstack/_internal/cli/services/presets/session.py b/src/dstack/_internal/cli/services/presets/session.py
index 72ec86e0fe..1f32bea6da 100644
--- a/src/dstack/_internal/cli/services/presets/session.py
+++ b/src/dstack/_internal/cli/services/presets/session.py
@@ -93,7 +93,7 @@ def write_agent_info(self, auth: "ClaudeAuth") -> None:
             _get_claude_version,
         )
 
-        info = ClaudeAgentInfo.parse_obj(
+        info = ClaudeAgentInfo.model_validate(
             {
                 "executable": auth.executable,
                 "version": _get_claude_version(auth),
@@ -105,7 +105,8 @@ def write_agent_info(self, auth: "ClaudeAuth") -> None:
             }
         )
         _write_private_text(
-            self.path / "agent.json", json.dumps(json.loads(info.json()), indent=2) + "\n"
+            self.path / "agent.json",
+            json.dumps(json.loads(info.model_dump_json()), indent=2) + "\n",
         )
 
     def append_log(self, line: str) -> None:
@@ -177,7 +178,7 @@ def create_preset_agent_session(
             "debug": debug,
         }
         _write_private_text(path / _SESSION_FILENAME, json.dumps(manifest, indent=2) + "\n")
-        data = json.loads(configuration.json(exclude_none=True))
+        data = json.loads(configuration.model_dump_json(exclude_none=True))
         if configuration.env:
             data["env"] = list(configuration.env)
         else:
diff --git a/src/dstack/_internal/cli/services/presets/store.py b/src/dstack/_internal/cli/services/presets/store.py
index 6f944d71c2..b48b350e4a 100644
--- a/src/dstack/_internal/cli/services/presets/store.py
+++ b/src/dstack/_internal/cli/services/presets/store.py
@@ -96,7 +96,7 @@ def release_name(self, name: str) -> Preset | None:
         preset = self.find_by_name(name)
         if preset is None:
             return None
-        detached = preset.copy(update={"name": None})
+        detached = preset.model_copy(update={"name": None})
         self.save(detached)
         return detached
 
@@ -138,7 +138,7 @@ def _migrate_legacy(self) -> None:
     def _load(self, path: Path) -> Preset:
         try:
             with path.open(encoding="utf-8") as f:
-                return Preset.parse_obj(yaml.safe_load(f))
+                return Preset.model_validate(yaml.safe_load(f))
         except (OSError, ValidationError, yaml.YAMLError) as e:
             raise CLIError(f"Invalid preset file {path}: {e}") from e
 
@@ -167,7 +167,7 @@ def _parse_preset_configuration(stream: TextIO) -> PresetConfiguration:
         data = yaml.safe_load(stream)
         if not isinstance(data, dict):
             raise ConfigurationError("Preset configuration must be a YAML object")
-        configuration = PresetConfiguration.parse_obj(data)
+        configuration = PresetConfiguration.model_validate(data)
     except ValidationError as e:
         raise ConfigurationError(e) from e
     except yaml.YAMLError as e:
diff --git a/src/dstack/_internal/cli/services/presets/verify.py b/src/dstack/_internal/cli/services/presets/verify.py
index 983b397a4c..2dd4c8005c 100644
--- a/src/dstack/_internal/cli/services/presets/verify.py
+++ b/src/dstack/_internal/cli/services/presets/verify.py
@@ -52,7 +52,7 @@ def load_preset_agent_report(
     # check below still rejects unknown leaked tokens.
     report_data = redact_structure(report_data, redacted_values)
     try:
-        report = AgentFinalReport.parse_obj(report_data)
+        report = AgentFinalReport.model_validate(report_data)
     except ValidationError as e:
         raise CLIError(f"Claude returned an invalid final report: {e}") from e
     if not report.success:
@@ -100,13 +100,13 @@ def build_verified_preset(
     target_type = (
         "gateway" if urlparse(run.service.url).scheme in {"http", "https"} else "server-proxy"
     )
-    benchmark = report.benchmark.copy(
+    benchmark = report.benchmark.model_copy(
         update={
             "target": PresetBenchmarkTarget(type=target_type),
             "client": PresetBenchmarkClient(type="local"),
         }
     )
-    portable_service = service.copy(deep=True)
+    portable_service = service.model_copy(deep=True)
     # The CLI resolved preset env references before submission; presets retain the references.
     for key, value in preset_configuration.env.items():
         if isinstance(value, EnvSentinel) and key in portable_service.env:
diff --git a/src/dstack/_internal/cli/services/profile.py b/src/dstack/_internal/cli/services/profile.py
index 0ef420d77e..3daae28225 100644
--- a/src/dstack/_internal/cli/services/profile.py
+++ b/src/dstack/_internal/cli/services/profile.py
@@ -2,14 +2,18 @@
 import os
 
 from dstack._internal.core.errors import CLIError
+from dstack._internal.core.models.backends.base import BackendType
+from dstack._internal.core.models.duration import (
+    parse_duration,
+    parse_idle_duration,
+    parse_off_duration,
+)
 from dstack._internal.core.models.profiles import (
     CreationPolicy,
     Profile,
     ProfileParams,
     ProfileRetry,
     SpotPolicy,
-    parse_duration,
-    parse_max_duration,
 )
 from dstack._internal.utils.env import environ
 from dstack._internal.utils.path import PathLike
@@ -61,6 +65,7 @@ def register_profile_args(parser: argparse.ArgumentParser):
         action="append",
         metavar="NAME",
         dest="backends",
+        type=BackendType,
         help="The backends that will be tried for provisioning",
     )
     profile_group.add_argument(
@@ -104,8 +109,9 @@ def register_profile_args(parser: argparse.ArgumentParser):
     fleets_group_exc.add_argument(
         "--idle-duration",
         dest="idle_duration",
-        type=str,
+        type=idle_duration,
         help="Time to wait before destroying the idle instance (if the run provisions a new instance)",
+        metavar="DURATION",
     )
 
     spot_group = parser.add_argument_group("Spot policy")
@@ -204,8 +210,12 @@ def apply_profile_args(
         )
 
 
-def max_duration(v: str) -> int:
-    return parse_max_duration(v)
+def max_duration(v: str):
+    return parse_off_duration(v)
+
+
+def idle_duration(v: str):
+    return parse_idle_duration(v)
 
 
 def retry_duration(v: str) -> int:
diff --git a/src/dstack/_internal/cli/services/resources.py b/src/dstack/_internal/cli/services/resources.py
index e81b6078db..7d69db0b3e 100644
--- a/src/dstack/_internal/cli/services/resources.py
+++ b/src/dstack/_internal/cli/services/resources.py
@@ -45,9 +45,9 @@ def register_resources_args(parser: ArgsParser) -> None:
 
 def apply_resources_args(args: argparse.Namespace, conf: AnyRunConfiguration) -> None:
     if args.cpu_spec:
-        conf.resources.cpu = resources.CPUSpec.parse_obj(args.cpu_spec)
+        conf.resources.cpu = resources.CPUSpec.model_validate(args.cpu_spec)
     if args.gpu_spec:
-        conf.resources.gpu = resources.GPUSpec.parse_obj(args.gpu_spec)
+        conf.resources.gpu = resources.GPUSpec.model_validate(args.gpu_spec)
     if args.memory_spec:
         conf.resources.memory = args.memory_spec
     if args.disk_spec:
diff --git a/src/dstack/_internal/cli/utils/gateway.py b/src/dstack/_internal/cli/utils/gateway.py
index 4c80aaaa8c..704741c89c 100644
--- a/src/dstack/_internal/cli/utils/gateway.py
+++ b/src/dstack/_internal/cli/utils/gateway.py
@@ -53,7 +53,7 @@ def print_gateways_json(gateways: List[Gateway], project: str) -> None:
         project=project,
         gateways=gateways,
     )
-    print(output.json())
+    print(output.model_dump_json())
 
 
 def get_gateways_table(
diff --git a/src/dstack/_internal/cli/utils/gpu.py b/src/dstack/_internal/cli/utils/gpu.py
index 3d19b173ba..7eb0fc9bb5 100644
--- a/src/dstack/_internal/cli/utils/gpu.py
+++ b/src/dstack/_internal/cli/utils/gpu.py
@@ -31,7 +31,7 @@ def print_gpu_json(
         gpus=gpus,
     )
 
-    print(output.json())
+    print(output.model_dump_json())
 
 
 def print_gpu_table(gpus: List[GpuGroup], run_spec: RunSpec, group_by: List[str], project: str):
diff --git a/src/dstack/_internal/cli/utils/run.py b/src/dstack/_internal/cli/utils/run.py
index fd90b35708..6c27f2aa6f 100644
--- a/src/dstack/_internal/cli/utils/run.py
+++ b/src/dstack/_internal/cli/utils/run.py
@@ -59,7 +59,7 @@ def print_runs_json(project: str, runs: List[Run]) -> None:
         project=project,
         runs=[r._run for r in runs],
     )
-    print(output.json())
+    print(output.model_dump_json())
 
 
 def print_run_plan(
diff --git a/src/dstack/_internal/core/backends/aws/compute.py b/src/dstack/_internal/core/backends/aws/compute.py
index 7fc586b4a7..2e0a65b568 100644
--- a/src/dstack/_internal/core/backends/aws/compute.py
+++ b/src/dstack/_internal/core/backends/aws/compute.py
@@ -55,7 +55,7 @@
     ProvisioningError,
 )
 from dstack._internal.core.models.backends.base import BackendType
-from dstack._internal.core.models.common import CoreModel
+from dstack._internal.core.models.common import CoreModel, validate_json_extra_ignore
 from dstack._internal.core.models.gateways import (
     GatewayComputeConfiguration,
     GatewayLoadBalancerConfiguration,
@@ -204,7 +204,7 @@ def get_offers_post_filter(
 
     def _get_offers_post_filter_cached_key(self, requirements: Requirements) -> int:
         # Requirements is not hashable, so we use a hack to get arguments hash
-        return hash(requirements.json())
+        return hash(requirements.model_dump_json())
 
     @cachedmethod(
         cache=lambda self: self._offers_post_filter_cache.cache,
@@ -468,7 +468,7 @@ def update_provisioning_data(
             )
             provisioning_data.backend_data = AWSInstanceBackendData(
                 eip_allocation_id=allocation_id
-            ).json()
+            ).model_dump_json()
             provisioning_data.hostname = public_ip
         else:
             provisioning_data.hostname = _get_instance_ip(
@@ -715,7 +715,7 @@ def create_gateway_load_balancer(
                 tg_arn=tg_arn,
                 listener_arn=listener_arn,
                 http_listener_arn=http_listener_arn,
-            ).json(),
+            ).model_dump_json(),
         )
 
     def terminate_gateway(
@@ -742,7 +742,7 @@ def terminate_gateway_load_balancer(
             )
             return
         try:
-            backend_data_parsed = AWSGatewayBackendData.__response__.parse_raw(backend_data)
+            backend_data_parsed = validate_json_extra_ignore(AWSGatewayBackendData, backend_data)
         except ValidationError:
             logger.exception(
                 "Failed to terminate load balancer for gateway %s: backend_data parsing error.",
@@ -772,8 +772,8 @@ def register_gateway_replica_with_load_balancer(
                 " gateway_backend_data is None"
             )
         try:
-            gateway_backend_data_parsed = AWSGatewayBackendData.__response__.parse_raw(
-                gateway_backend_data
+            gateway_backend_data_parsed = validate_json_extra_ignore(
+                AWSGatewayBackendData, gateway_backend_data
             )
         except ValidationError as e:
             raise ComputeError(
@@ -810,8 +810,8 @@ def deregister_gateway_replica_from_load_balancer(
                 " gateway_backend_data is None"
             )
         try:
-            gateway_backend_data_parsed = AWSGatewayBackendData.__response__.parse_raw(
-                gateway_backend_data
+            gateway_backend_data_parsed = validate_json_extra_ignore(
+                AWSGatewayBackendData, gateway_backend_data
             )
         except ValidationError as e:
             raise ComputeError(
@@ -861,7 +861,7 @@ def register_volume(self, volume: Volume) -> VolumeProvisioningData:
             backend_data=AWSVolumeBackendData(
                 volume_type=response_volume["VolumeType"],
                 iops=response_volume["Iops"],
-            ).json(),
+            ).model_dump_json(),
         )
 
     def create_volume(self, volume: Volume) -> VolumeProvisioningData:
@@ -920,7 +920,7 @@ def create_volume(self, volume: Volume) -> VolumeProvisioningData:
             backend_data=AWSVolumeBackendData(
                 volume_type=response["VolumeType"],
                 iops=iops,
-            ).json(),
+            ).model_dump_json(),
         )
 
     def delete_volume(self, volume: Volume):
@@ -1161,7 +1161,10 @@ def _get_image_id_and_username_cache_key(
         image_config: Optional[AWSOSImageConfig] = None,
     ) -> tuple:
         return hashkey(
-            region, gpu_name, instance_type, image_config.json() if image_config else None
+            region,
+            gpu_name,
+            instance_type,
+            image_config.model_dump_json() if image_config else None,
         )
 
     @cachedmethod(
@@ -1424,7 +1427,7 @@ def _parse_instance_backend_data(backend_data: Optional[str]) -> "AWSInstanceBac
     if backend_data is None:
         return AWSInstanceBackendData()
     try:
-        return AWSInstanceBackendData.__response__.parse_raw(backend_data)
+        return validate_json_extra_ignore(AWSInstanceBackendData, backend_data)
     except ValidationError:
         logger.exception("Failed to parse AWS instance backend_data; treating as empty")
         return AWSInstanceBackendData()
diff --git a/src/dstack/_internal/core/backends/aws/configurator.py b/src/dstack/_internal/core/backends/aws/configurator.py
index a7a1b92a3a..2920ac1c97 100644
--- a/src/dstack/_internal/core/backends/aws/configurator.py
+++ b/src/dstack/_internal/core/backends/aws/configurator.py
@@ -28,6 +28,10 @@
 from dstack._internal.core.models.backends.base import (
     BackendType,
 )
+from dstack._internal.core.models.common import (
+    validate_extra_ignore,
+    validate_json_extra_ignore,
+)
 from dstack._internal.utils.logging import get_logger
 
 logger = get_logger(__name__)
@@ -86,27 +90,30 @@ def create_backend(
             config.regions = DEFAULT_REGIONS
         return BackendRecord(
             config=AWSStoredConfig(
-                **AWSBackendConfig.__response__.parse_obj(config).dict()
-            ).json(),
-            auth=AWSCreds.parse_obj(config.creds).json(),
+                **validate_extra_ignore(AWSBackendConfig, config).model_dump()
+            ).model_dump_json(),
+            auth=AWSCreds.model_validate(config.creds).model_dump_json(),
         )
 
     def get_backend_config_with_creds(self, record: BackendRecord) -> AWSBackendConfigWithCreds:
         config = self._get_config(record)
-        return AWSBackendConfigWithCreds.__response__.parse_obj(config)
+        return validate_extra_ignore(AWSBackendConfigWithCreds, config)
 
     def get_backend_config_without_creds(self, record: BackendRecord) -> AWSBackendConfig:
         config = self._get_config(record)
-        return AWSBackendConfig.__response__.parse_obj(config)
+        return validate_extra_ignore(AWSBackendConfig, config)
 
     def get_backend(self, record: BackendRecord) -> AWSBackend:
         config = self._get_config(record)
         return AWSBackend(config=config)
 
     def _get_config(self, record: BackendRecord) -> AWSConfig:
-        return AWSConfig.__response__(
-            **json.loads(record.config),
-            creds=AWSCreds.__response__.parse_raw(record.auth).__root__,
+        return validate_extra_ignore(
+            AWSConfig,
+            {
+                **json.loads(record.config),
+                "creds": validate_json_extra_ignore(AWSCreds, record.auth).root,
+            },
         )
 
     def _check_config_tags(self, config: AWSBackendConfigWithCreds):
@@ -179,7 +186,8 @@ def _check_config_vpc(self, session: Session, config: AWSBackendConfigWithCreds)
                 future = executor.submit(
                     compute.get_vpc_id_subnets_ids_or_error,
                     ec2_client=ec2_client,
-                    config=AWSConfig.parse_obj(config),
+                    # `config` is an `AWSBackendConfigWithCreds`, a different class.
+                    config=AWSConfig.model_validate(config.model_dump()),
                     region=region,
                     allocate_public_ip=allocate_public_ip,
                 )
diff --git a/src/dstack/_internal/core/backends/aws/models.py b/src/dstack/_internal/core/backends/aws/models.py
index 11b9708d00..2dce3bcecc 100644
--- a/src/dstack/_internal/core/backends/aws/models.py
+++ b/src/dstack/_internal/core/backends/aws/models.py
@@ -1,6 +1,6 @@
 from typing import Annotated, Dict, List, Literal, Optional, Union
 
-from pydantic import Field
+from pydantic import Field, RootModel
 
 from dstack._internal.core.models.common import CoreModel
 
@@ -9,7 +9,7 @@ class AWSOSImage(CoreModel):
     name: Annotated[str, Field(description="The AMI name")]
     owner: Annotated[
         str,
-        Field(regex=r"^(\d{12}|self)$", description="The AMI owner, account ID or `self`"),
+        Field(pattern=r"^(\d{12}|self)$", description="The AMI owner, account ID or `self`"),
     ] = "self"
     user: Annotated[str, Field(description="The OS user for provisioning")]
 
@@ -38,8 +38,8 @@ class AWSDefaultCreds(CoreModel):
 AnyAWSCreds = Union[AWSAccessKeyCreds, AWSDefaultCreds]
 
 
-class AWSCreds(CoreModel):
-    __root__: AnyAWSCreds = Field(..., discriminator="type")
+class AWSCreds(RootModel[Annotated[AnyAWSCreds, Field(discriminator="type")]]):
+    pass
 
 
 class AWSBackendConfig(CoreModel):
diff --git a/src/dstack/_internal/core/backends/azure/configurator.py b/src/dstack/_internal/core/backends/azure/configurator.py
index 33a8576981..bea1b3bad3 100644
--- a/src/dstack/_internal/core/backends/azure/configurator.py
+++ b/src/dstack/_internal/core/backends/azure/configurator.py
@@ -46,6 +46,7 @@
 from dstack._internal.core.models.backends.base import (
     BackendType,
 )
+from dstack._internal.core.models.common import validate_extra_ignore, validate_json_extra_ignore
 
 LOCATIONS = [
     ("(US) Central US", "centralus"),
@@ -129,18 +130,18 @@ def create_backend(
         )
         return BackendRecord(
             config=AzureStoredConfig(
-                **AzureBackendConfig.__response__.parse_obj(config).dict()
-            ).json(),
-            auth=AzureCreds.parse_obj(config.creds).__root__.json(),
+                **validate_extra_ignore(AzureBackendConfig, config).model_dump()
+            ).model_dump_json(),
+            auth=AzureCreds.model_validate(config.creds).root.model_dump_json(),
         )
 
     def get_backend_config_with_creds(self, record: BackendRecord) -> AzureBackendConfigWithCreds:
         config = self._get_config(record)
-        return AzureBackendConfigWithCreds.__response__.parse_obj(config)
+        return validate_extra_ignore(AzureBackendConfigWithCreds, config)
 
     def get_backend_config_without_creds(self, record: BackendRecord) -> AzureBackendConfig:
         config = self._get_config(record)
-        return AzureBackendConfig.__response__.parse_obj(config)
+        return validate_extra_ignore(AzureBackendConfig, config)
 
     def get_backend(self, record: BackendRecord) -> AzureBackend:
         config = self._get_config(record)
@@ -152,10 +153,13 @@ def _get_config(self, record: BackendRecord) -> AzureConfig:
         if regions is None:
             # Legacy config stores regions as locations
             regions = config_dict.pop("locations")
-        return AzureConfig.__response__(
-            **config_dict,
-            regions=regions,
-            creds=AzureCreds.__response__.parse_raw(record.auth).__root__,
+        return validate_extra_ignore(
+            AzureConfig,
+            {
+                **config_dict,
+                "regions": regions,
+                "creds": validate_json_extra_ignore(AzureCreds, record.auth).root,
+            },
         )
 
     def _check_config_tenant_id(
diff --git a/src/dstack/_internal/core/backends/azure/models.py b/src/dstack/_internal/core/backends/azure/models.py
index 0d7c11a116..dc6c174c34 100644
--- a/src/dstack/_internal/core/backends/azure/models.py
+++ b/src/dstack/_internal/core/backends/azure/models.py
@@ -1,6 +1,6 @@
 from typing import Annotated, Dict, List, Literal, Optional, Union
 
-from pydantic import Field
+from pydantic import Field, RootModel
 
 from dstack._internal.core.models.common import CoreModel
 
@@ -20,8 +20,8 @@ class AzureDefaultCreds(CoreModel):
 AnyAzureCreds = Union[AzureClientCreds, AzureDefaultCreds]
 
 
-class AzureCreds(CoreModel):
-    __root__: AnyAzureCreds = Field(..., discriminator="type")
+class AzureCreds(RootModel[Annotated[AnyAzureCreds, Field(discriminator="type")]]):
+    pass
 
 
 class AzureBackendConfig(CoreModel):
diff --git a/src/dstack/_internal/core/backends/base/compute.py b/src/dstack/_internal/core/backends/base/compute.py
index 33013abdf8..cf0e6a919a 100644
--- a/src/dstack/_internal/core/backends/base/compute.py
+++ b/src/dstack/_internal/core/backends/base/compute.py
@@ -342,7 +342,7 @@ def _get_offers_cached_key(
     ) -> int:
         hash_items: list[Union[str, bool]] = []
         # Requirements is not hashable, so we use a hack to get arguments hash
-        hash_items.append(requirements.json())
+        hash_items.append(requirements.model_dump_json())
         if self.full_offers_argument_has_effect:
             hash_items.append(full_offers)
         if self.unallocated_resources_argument_has_effect:
@@ -405,7 +405,7 @@ def run_job(
             reservation=job.job_spec.requirements.reservation,
             tags=run.run_spec.merged_profile.tags,
         )
-        instance_offer = instance_offer.copy()
+        instance_offer = instance_offer.model_copy()
         self._restrict_instance_offer_az_to_volumes_az(instance_offer, volumes)
         return self.create_instance(
             instance_offer, instance_config, placement_group=placement_group
diff --git a/src/dstack/_internal/core/backends/base/offers.py b/src/dstack/_internal/core/backends/base/offers.py
index 681b6f9557..cb3f974f87 100644
--- a/src/dstack/_internal/core/backends/base/offers.py
+++ b/src/dstack/_internal/core/backends/base/offers.py
@@ -6,7 +6,6 @@
 
 import gpuhunt
 from cachetools import TTLCache
-from pydantic import parse_obj_as
 
 from dstack._internal.core.models.backends.base import BackendType
 from dstack._internal.core.models.instances import (
@@ -106,17 +105,17 @@ def catalog_item_to_offer(
     )
     if disk_size_mib is None:
         return None
-    resources = Resources.construct(
+    resources = Resources.model_construct(
         cpu_arch=item.cpu_arch,
         cpus=item.cpu,
         memory_mib=round(item.memory * 1024),
         gpus=gpus,
         spot=item.spot,
-        disk=Disk.construct(size_mib=disk_size_mib),
+        disk=Disk.model_construct(size_mib=disk_size_mib),
     )
-    return InstanceOffer.construct(
+    return InstanceOffer.model_construct(
         backend=backend,
-        instance=InstanceType.construct(
+        instance=InstanceType.model_construct(
             name=item.instance_name,
             resources=resources,
         ),
@@ -172,7 +171,7 @@ def requirements_to_query_filter(req: Optional[Requirements]) -> gpuhunt.QueryFi
     res = req.resources
     if res.cpu:
         # TODO: Remove in 0.20. Use res.cpu directly
-        cpu = parse_obj_as(CPUSpec, res.cpu)
+        cpu = CPUSpec.model_validate(res.cpu)
         q.cpu_arch = cpu.arch
         q.min_cpu = cpu.count.min
         q.max_cpu = cpu.count.max
@@ -250,7 +249,7 @@ def modifier(offer: InstanceOfferWithAvailability) -> Optional[InstanceOfferWith
         disk_size_range = requirements_disk_range.intersect(configurable_disk_size)
         if disk_size_range is None:
             return None
-        offer_copy = offer.copy(deep=True)
+        offer_copy = offer.model_copy(deep=True)
         offer_copy.instance.resources.disk = Disk(
             size_mib=get_or_error(disk_size_range.min) * 1024
         )
diff --git a/src/dstack/_internal/core/backends/cloudrift/configurator.py b/src/dstack/_internal/core/backends/cloudrift/configurator.py
index 9f69958026..82dd1ebc64 100644
--- a/src/dstack/_internal/core/backends/cloudrift/configurator.py
+++ b/src/dstack/_internal/core/backends/cloudrift/configurator.py
@@ -18,6 +18,7 @@
 from dstack._internal.core.models.backends.base import (
     BackendType,
 )
+from dstack._internal.core.models.common import validate_extra_ignore, validate_json_extra_ignore
 
 
 class CloudRiftConfigurator(
@@ -39,29 +40,32 @@ def create_backend(
     ) -> BackendRecord:
         return BackendRecord(
             config=CloudRiftStoredConfig(
-                **CloudRiftBackendConfig.__response__.parse_obj(config).dict()
-            ).json(),
-            auth=CloudRiftCreds.parse_obj(config.creds).json(),
+                **validate_extra_ignore(CloudRiftBackendConfig, config).model_dump()
+            ).model_dump_json(),
+            auth=CloudRiftCreds.model_validate(config.creds).model_dump_json(),
         )
 
     def get_backend_config_with_creds(
         self, record: BackendRecord
     ) -> CloudRiftBackendConfigWithCreds:
         config = self._get_config(record)
-        return CloudRiftBackendConfigWithCreds.__response__.parse_obj(config)
+        return validate_extra_ignore(CloudRiftBackendConfigWithCreds, config)
 
     def get_backend_config_without_creds(self, record: BackendRecord) -> CloudRiftBackendConfig:
         config = self._get_config(record)
-        return CloudRiftBackendConfig.__response__.parse_obj(config)
+        return validate_extra_ignore(CloudRiftBackendConfig, config)
 
     def get_backend(self, record: BackendRecord) -> CloudRiftBackend:
         config = self._get_config(record)
         return CloudRiftBackend(config=config)
 
     def _get_config(self, record: BackendRecord) -> CloudRiftConfig:
-        return CloudRiftConfig.__response__(
-            **json.loads(record.config),
-            creds=CloudRiftCreds.__response__.parse_raw(record.auth),
+        return validate_extra_ignore(
+            CloudRiftConfig,
+            {
+                **json.loads(record.config),
+                "creds": validate_json_extra_ignore(CloudRiftCreds, record.auth),
+            },
         )
 
     def _validate_creds(self, creds: AnyCloudRiftCreds):
diff --git a/src/dstack/_internal/core/backends/crusoe/compute.py b/src/dstack/_internal/core/backends/crusoe/compute.py
index 97b3b85a31..13e54eef67 100644
--- a/src/dstack/_internal/core/backends/crusoe/compute.py
+++ b/src/dstack/_internal/core/backends/crusoe/compute.py
@@ -24,7 +24,7 @@
 from dstack._internal.core.backends.crusoe.resources import CrusoeClient
 from dstack._internal.core.errors import BackendError, NotYetTerminated
 from dstack._internal.core.models.backends.base import BackendType
-from dstack._internal.core.models.common import CoreModel
+from dstack._internal.core.models.common import CoreModel, validate_json_extra_ignore
 from dstack._internal.core.models.instances import (
     InstanceAvailability,
     InstanceConfiguration,
@@ -291,7 +291,7 @@ def create_instance(
             ssh_port=22,
             username="ubuntu",
             dockerized=True,
-            backend_data=CrusoeInstanceBackendData(data_disk_id=data_disk_id).json(),
+            backend_data=CrusoeInstanceBackendData(data_disk_id=data_disk_id).model_dump_json(),
         )
 
     def update_provisioning_data(
@@ -356,7 +356,7 @@ def create_placement_group(
                 backend=BackendType.CRUSOE,
                 backend_data=CrusoePlacementGroupBackendData(
                     ib_partition_id=None, ib_network_id=None
-                ).json(),
+                ).model_dump_json(),
             )
 
         ib_networks = self._client.list_ib_networks()
@@ -385,7 +385,7 @@ def create_placement_group(
             backend_data=CrusoePlacementGroupBackendData(
                 ib_partition_id=partition["id"],
                 ib_network_id=target_network["id"],
-            ).json(),
+            ).model_dump_json(),
         )
 
     def delete_placement_group(self, placement_group: PlacementGroup) -> None:
@@ -422,7 +422,7 @@ class CrusoeInstanceBackendData(CoreModel):
     def load(cls, raw: Optional[str]) -> "CrusoeInstanceBackendData":
         if raw is None:
             return cls()
-        return cls.__response__.parse_raw(raw)
+        return validate_json_extra_ignore(cls, raw)
 
 
 class CrusoePlacementGroupBackendData(CoreModel):
@@ -433,4 +433,4 @@ class CrusoePlacementGroupBackendData(CoreModel):
     def load(cls, raw: Optional[str]) -> "CrusoePlacementGroupBackendData":
         if raw is None:
             return cls()
-        return cls.__response__.parse_raw(raw)
+        return validate_json_extra_ignore(cls, raw)
diff --git a/src/dstack/_internal/core/backends/crusoe/configurator.py b/src/dstack/_internal/core/backends/crusoe/configurator.py
index a6e4274a43..d10f4dc535 100644
--- a/src/dstack/_internal/core/backends/crusoe/configurator.py
+++ b/src/dstack/_internal/core/backends/crusoe/configurator.py
@@ -15,6 +15,7 @@
 )
 from dstack._internal.core.backends.crusoe.resources import CrusoeClient
 from dstack._internal.core.models.backends.base import BackendType
+from dstack._internal.core.models.common import validate_extra_ignore, validate_json_extra_ignore
 
 
 class CrusoeConfigurator(
@@ -54,25 +55,28 @@ def create_backend(
     ) -> BackendRecord:
         return BackendRecord(
             config=CrusoeStoredConfig(
-                **CrusoeBackendConfig.__response__.parse_obj(config).dict()
-            ).json(),
-            auth=CrusoeCreds.parse_obj(config.creds).json(),
+                **validate_extra_ignore(CrusoeBackendConfig, config).model_dump()
+            ).model_dump_json(),
+            auth=CrusoeCreds.model_validate(config.creds).model_dump_json(),
         )
 
     def get_backend_config_with_creds(self, record: BackendRecord) -> CrusoeBackendConfigWithCreds:
         config = self._get_config(record)
-        return CrusoeBackendConfigWithCreds.__response__.parse_obj(config)
+        return validate_extra_ignore(CrusoeBackendConfigWithCreds, config)
 
     def get_backend_config_without_creds(self, record: BackendRecord) -> CrusoeBackendConfig:
         config = self._get_config(record)
-        return CrusoeBackendConfig.__response__.parse_obj(config)
+        return validate_extra_ignore(CrusoeBackendConfig, config)
 
     def get_backend(self, record: BackendRecord) -> CrusoeBackend:
         config = self._get_config(record)
         return CrusoeBackend(config=config)
 
     def _get_config(self, record: BackendRecord) -> CrusoeConfig:
-        return CrusoeConfig.__response__(
-            **json.loads(record.config),
-            creds=CrusoeCreds.__response__.parse_raw(record.auth),
+        return validate_extra_ignore(
+            CrusoeConfig,
+            {
+                **json.loads(record.config),
+                "creds": validate_json_extra_ignore(CrusoeCreds, record.auth),
+            },
         )
diff --git a/src/dstack/_internal/core/backends/digitalocean_base/configurator.py b/src/dstack/_internal/core/backends/digitalocean_base/configurator.py
index 9f0fc21699..a20c128c5b 100644
--- a/src/dstack/_internal/core/backends/digitalocean_base/configurator.py
+++ b/src/dstack/_internal/core/backends/digitalocean_base/configurator.py
@@ -14,6 +14,7 @@
     BaseDigitalOceanCreds,
     BaseDigitalOceanStoredConfig,
 )
+from dstack._internal.core.models.common import validate_extra_ignore, validate_json_extra_ignore
 
 
 class BaseDigitalOceanConfigurator(Configurator):
@@ -27,30 +28,33 @@ def create_backend(
     ) -> BackendRecord:
         return BackendRecord(
             config=BaseDigitalOceanStoredConfig(
-                **BaseDigitalOceanBackendConfig.__response__.parse_obj(config).dict()
-            ).json(),
-            auth=BaseDigitalOceanCreds.parse_obj(config.creds).json(),
+                **validate_extra_ignore(BaseDigitalOceanBackendConfig, config).model_dump()
+            ).model_dump_json(),
+            auth=BaseDigitalOceanCreds.model_validate(config.creds).model_dump_json(),
         )
 
     def get_backend_config_with_creds(
         self, record: BackendRecord
     ) -> BaseDigitalOceanBackendConfigWithCreds:
         config = self._get_config(record)
-        return BaseDigitalOceanBackendConfigWithCreds.__response__.parse_obj(config)
+        return validate_extra_ignore(BaseDigitalOceanBackendConfigWithCreds, config)
 
     def get_backend_config_without_creds(
         self, record: BackendRecord
     ) -> BaseDigitalOceanBackendConfig:
         config = self._get_config(record)
-        return BaseDigitalOceanBackendConfig.__response__.parse_obj(config)
+        return validate_extra_ignore(BaseDigitalOceanBackendConfig, config)
 
     def get_backend(self, record: BackendRecord) -> BaseDigitalOceanBackend:
         raise NotImplementedError("Subclasses must implement get_backend")
 
     def _get_config(self, record: BackendRecord) -> BaseDigitalOceanConfig:
-        return BaseDigitalOceanConfig.__response__(
-            **json.loads(record.config),
-            creds=BaseDigitalOceanCreds.__response__.parse_raw(record.auth),
+        return validate_extra_ignore(
+            BaseDigitalOceanConfig,
+            {
+                **json.loads(record.config),
+                "creds": validate_json_extra_ignore(BaseDigitalOceanCreds, record.auth),
+            },
         )
 
     def _validate_creds(self, creds: AnyBaseDigitalOceanCreds, project_name: Optional[str] = None):
diff --git a/src/dstack/_internal/core/backends/gcp/compute.py b/src/dstack/_internal/core/backends/gcp/compute.py
index 54aa640bfb..b79857754d 100644
--- a/src/dstack/_internal/core/backends/gcp/compute.py
+++ b/src/dstack/_internal/core/backends/gcp/compute.py
@@ -55,7 +55,7 @@
     ProvisioningError,
 )
 from dstack._internal.core.models.backends.base import BackendType
-from dstack._internal.core.models.common import CoreModel
+from dstack._internal.core.models.common import CoreModel, validate_extra_ignore
 from dstack._internal.core.models.gateways import (
     GatewayComputeConfiguration,
     GatewayProvisioningData,
@@ -194,7 +194,7 @@ def reservation_modifier(
                             zones_with_capacity.append(zone)
                 if not matching_zones:
                     return None
-                offer = offer.copy(deep=True)
+                offer = offer.model_copy(deep=True)
                 if zones_with_capacity:
                     offer.availability_zones = zones_with_capacity
                 else:
@@ -214,8 +214,8 @@ def get_offers_post_filter(
 
             def reserved_offers_filter(offer: InstanceOfferWithAvailability) -> bool:
                 """Remove reserved-only offers"""
-                if GCPOfferBackendData.__response__.parse_obj(
-                    offer.backend_data
+                if validate_extra_ignore(
+                    GCPOfferBackendData, offer.backend_data
                 ).is_dws_calendar_mode:
                     return False
                 return True
@@ -678,7 +678,7 @@ def register_volume(self, volume: Volume) -> VolumeProvisioningData:
                     detachable=True,
                     backend_data=GCPVolumeDiskBackendData(
                         disk_type=gcp_resources.full_resource_name_to_name(disk.type_),
-                    ).json(),
+                    ).model_dump_json(),
                 )
         raise ComputeError(f"Persistent disk {volume.configuration.volume_id} not found")
 
@@ -746,7 +746,7 @@ def create_volume(self, volume: Volume) -> VolumeProvisioningData:
             detachable=True,
             backend_data=GCPVolumeDiskBackendData(
                 disk_type=gcp_resources.full_resource_name_to_name(disk.type_),
-            ).json(),
+            ).model_dump_json(),
         )
 
     def delete_volume(self, volume: Volume):
diff --git a/src/dstack/_internal/core/backends/gcp/configurator.py b/src/dstack/_internal/core/backends/gcp/configurator.py
index ce15264fd1..1621756cc9 100644
--- a/src/dstack/_internal/core/backends/gcp/configurator.py
+++ b/src/dstack/_internal/core/backends/gcp/configurator.py
@@ -23,6 +23,7 @@
 from dstack._internal.core.models.backends.base import (
     BackendType,
 )
+from dstack._internal.core.models.common import validate_extra_ignore, validate_json_extra_ignore
 
 LOCATIONS = [
     {
@@ -146,27 +147,30 @@ def create_backend(
             config.regions = DEFAULT_REGIONS
         return BackendRecord(
             config=GCPStoredConfig(
-                **GCPBackendConfig.__response__.parse_obj(config).dict(),
-            ).json(),
-            auth=GCPCreds.parse_obj(config.creds).json(),
+                **validate_extra_ignore(GCPBackendConfig, config).model_dump(),
+            ).model_dump_json(),
+            auth=GCPCreds.model_validate(config.creds).model_dump_json(),
         )
 
     def get_backend_config_with_creds(self, record: BackendRecord) -> GCPBackendConfigWithCreds:
         config = self._get_config(record)
-        return GCPBackendConfigWithCreds.__response__.parse_obj(config)
+        return validate_extra_ignore(GCPBackendConfigWithCreds, config)
 
     def get_backend_config_without_creds(self, record: BackendRecord) -> GCPBackendConfig:
         config = self._get_config(record)
-        return GCPBackendConfig.__response__.parse_obj(config)
+        return validate_extra_ignore(GCPBackendConfig, config)
 
     def get_backend(self, record: BackendRecord) -> GCPBackend:
         config = self._get_config(record)
         return GCPBackend(config=config)
 
     def _get_config(self, record: BackendRecord) -> GCPConfig:
-        return GCPConfig.__response__(
-            **json.loads(record.config),
-            creds=GCPCreds.__response__.parse_raw(record.auth).__root__,
+        return validate_extra_ignore(
+            GCPConfig,
+            {
+                **json.loads(record.config),
+                "creds": validate_json_extra_ignore(GCPCreds, record.auth).root,
+            },
         )
 
     def _check_config_tags(self, config: GCPBackendConfigWithCreds):
diff --git a/src/dstack/_internal/core/backends/gcp/models.py b/src/dstack/_internal/core/backends/gcp/models.py
index 10c5a42d0f..47809329d9 100644
--- a/src/dstack/_internal/core/backends/gcp/models.py
+++ b/src/dstack/_internal/core/backends/gcp/models.py
@@ -1,6 +1,6 @@
 from typing import Annotated, Dict, List, Literal, Optional, Union
 
-from pydantic import Field, root_validator
+from pydantic import Field, RootModel, model_validator
 
 from dstack._internal.core.backends.base.models import fill_data
 from dstack._internal.core.models.common import CoreModel
@@ -23,8 +23,8 @@ class GCPDefaultCreds(CoreModel):
 AnyGCPCreds = Union[GCPServiceAccountCreds, GCPDefaultCreds]
 
 
-class GCPCreds(CoreModel):
-    __root__: AnyGCPCreds = Field(..., discriminator="type")
+class GCPCreds(RootModel[Annotated[AnyGCPCreds, Field(discriminator="type")]]):
+    pass
 
 
 class GCPBackendConfig(CoreModel):
@@ -56,7 +56,7 @@ class GCPBackendConfig(CoreModel):
                 " A VPC should have eight subnets to maximize the bandwidth in clusters"
                 " with eight-GPU instances."
             ),
-            max_items=1,  # The currently supported instance types only need one VPC with eight subnets.
+            max_length=1,  # The currently supported instance types only need one VPC with eight subnets.
         ),
     ] = None
     vpc_project_id: Annotated[
@@ -96,7 +96,7 @@ class GCPBackendConfig(CoreModel):
                 "The list of preview GCP features to enable."
                 " There are currently no preview features"
             ),
-            max_items=1,
+            max_length=1,
         ),
     ] = None
 
@@ -121,7 +121,8 @@ class GCPServiceAccountFileCreds(CoreModel):
         ),
     ] = None
 
-    @root_validator
+    @model_validator(mode="before")
+    @classmethod
     def fill_data(cls, values):
         return fill_data(values)
 
diff --git a/src/dstack/_internal/core/backends/hotaisle/compute.py b/src/dstack/_internal/core/backends/hotaisle/compute.py
index c23ab717ce..eabe705ec6 100644
--- a/src/dstack/_internal/core/backends/hotaisle/compute.py
+++ b/src/dstack/_internal/core/backends/hotaisle/compute.py
@@ -19,7 +19,11 @@
 from dstack._internal.core.backends.hotaisle.api_client import HotAisleAPIClient
 from dstack._internal.core.backends.hotaisle.models import HotAisleConfig
 from dstack._internal.core.models.backends.base import BackendType
-from dstack._internal.core.models.common import CoreModel
+from dstack._internal.core.models.common import (
+    CoreModel,
+    validate_extra_ignore,
+    validate_json_extra_ignore,
+)
 from dstack._internal.core.models.instances import (
     InstanceAvailability,
     InstanceConfiguration,
@@ -74,8 +78,8 @@ def create_instance(
     ) -> JobProvisioningData:
         project_ssh_key = instance_config.ssh_keys[0]
         self.api_client.upload_ssh_key(project_ssh_key.public)
-        offer_backend_data: HotAisleOfferBackendData = (
-            HotAisleOfferBackendData.__response__.parse_obj(instance_offer.backend_data)
+        offer_backend_data: HotAisleOfferBackendData = validate_extra_ignore(
+            HotAisleOfferBackendData, instance_offer.backend_data
         )
         vm_data = self.api_client.create_virtual_machine(offer_backend_data.vm_specs)
         return JobProvisioningData(
@@ -90,7 +94,9 @@ def create_instance(
             ssh_port=22,
             dockerized=True,
             ssh_proxy=None,
-            backend_data=HotAisleInstanceBackendData(ip_address=vm_data["ip_address"]).json(),
+            backend_data=HotAisleInstanceBackendData(
+                ip_address=vm_data["ip_address"]
+            ).model_dump_json(),
         )
 
     def update_provisioning_data(
@@ -182,7 +188,7 @@ class HotAisleInstanceBackendData(CoreModel):
     @classmethod
     def load(cls, raw: Optional[str]) -> "HotAisleInstanceBackendData":
         assert raw is not None
-        return cls.__response__.parse_raw(raw)
+        return validate_json_extra_ignore(cls, raw)
 
 
 class HotAisleOfferBackendData(CoreModel):
diff --git a/src/dstack/_internal/core/backends/hotaisle/configurator.py b/src/dstack/_internal/core/backends/hotaisle/configurator.py
index 19162ca992..6d3ff7a393 100644
--- a/src/dstack/_internal/core/backends/hotaisle/configurator.py
+++ b/src/dstack/_internal/core/backends/hotaisle/configurator.py
@@ -17,6 +17,7 @@
 from dstack._internal.core.models.backends.base import (
     BackendType,
 )
+from dstack._internal.core.models.common import validate_extra_ignore, validate_json_extra_ignore
 
 
 class HotAisleConfigurator(
@@ -36,29 +37,32 @@ def create_backend(
     ) -> BackendRecord:
         return BackendRecord(
             config=HotAisleStoredConfig(
-                **HotAisleBackendConfig.__response__.parse_obj(config).dict()
-            ).json(),
-            auth=HotAisleCreds.parse_obj(config.creds).json(),
+                **validate_extra_ignore(HotAisleBackendConfig, config).model_dump()
+            ).model_dump_json(),
+            auth=HotAisleCreds.model_validate(config.creds).model_dump_json(),
         )
 
     def get_backend_config_with_creds(
         self, record: BackendRecord
     ) -> HotAisleBackendConfigWithCreds:
         config = self._get_config(record)
-        return HotAisleBackendConfigWithCreds.__response__.parse_obj(config)
+        return validate_extra_ignore(HotAisleBackendConfigWithCreds, config)
 
     def get_backend_config_without_creds(self, record: BackendRecord) -> HotAisleBackendConfig:
         config = self._get_config(record)
-        return HotAisleBackendConfig.__response__.parse_obj(config)
+        return validate_extra_ignore(HotAisleBackendConfig, config)
 
     def get_backend(self, record: BackendRecord) -> HotAisleBackend:
         config = self._get_config(record)
         return HotAisleBackend(config=config)
 
     def _get_config(self, record: BackendRecord) -> HotAisleConfig:
-        return HotAisleConfig.__response__(
-            **json.loads(record.config),
-            creds=HotAisleCreds.__response__.parse_raw(record.auth),
+        return validate_extra_ignore(
+            HotAisleConfig,
+            {
+                **json.loads(record.config),
+                "creds": validate_json_extra_ignore(HotAisleCreds, record.auth),
+            },
         )
 
     def _validate_creds(self, creds: AnyHotAisleCreds, team_handle: str):
diff --git a/src/dstack/_internal/core/backends/jarvislabs/compute.py b/src/dstack/_internal/core/backends/jarvislabs/compute.py
index f6468ab89f..fcdb65a9f1 100644
--- a/src/dstack/_internal/core/backends/jarvislabs/compute.py
+++ b/src/dstack/_internal/core/backends/jarvislabs/compute.py
@@ -26,7 +26,7 @@
 from dstack._internal.core.backends.jarvislabs.models import JarvisLabsConfig
 from dstack._internal.core.errors import ProvisioningError
 from dstack._internal.core.models.backends.base import BackendType
-from dstack._internal.core.models.common import CoreModel
+from dstack._internal.core.models.common import CoreModel, validate_json_extra_ignore
 from dstack._internal.core.models.instances import (
     InstanceAvailability,
     InstanceConfiguration,
@@ -62,7 +62,7 @@ class JarvisLabsInstanceBackendData(CoreModel):
     def load(cls, raw: Optional[str]) -> "JarvisLabsInstanceBackendData":
         if raw is None:
             return cls()
-        return cls.__response__.parse_raw(raw)
+        return validate_json_extra_ignore(cls, raw)
 
 
 class JarvisLabsCompute(
@@ -169,7 +169,7 @@ def create_instance(
             ssh_port=22,
             dockerized=True,
             ssh_proxy=None,
-            backend_data=JarvisLabsInstanceBackendData(ssh_key_ids=ssh_key_ids).json(),
+            backend_data=JarvisLabsInstanceBackendData(ssh_key_ids=ssh_key_ids).model_dump_json(),
         )
 
     def update_provisioning_data(
diff --git a/src/dstack/_internal/core/backends/jarvislabs/configurator.py b/src/dstack/_internal/core/backends/jarvislabs/configurator.py
index 041694256f..1dd358a92c 100644
--- a/src/dstack/_internal/core/backends/jarvislabs/configurator.py
+++ b/src/dstack/_internal/core/backends/jarvislabs/configurator.py
@@ -18,6 +18,7 @@
 )
 from dstack._internal.core.errors import ServerClientError
 from dstack._internal.core.models.backends.base import BackendType
+from dstack._internal.core.models.common import validate_extra_ignore, validate_json_extra_ignore
 
 
 class JarvisLabsConfigurator(
@@ -40,29 +41,32 @@ def create_backend(
     ) -> BackendRecord:
         return BackendRecord(
             config=JarvisLabsStoredConfig(
-                **JarvisLabsBackendConfig.__response__.parse_obj(config).dict()
-            ).json(),
-            auth=JarvisLabsCreds.parse_obj(config.creds).json(),
+                **validate_extra_ignore(JarvisLabsBackendConfig, config).model_dump()
+            ).model_dump_json(),
+            auth=JarvisLabsCreds.model_validate(config.creds).model_dump_json(),
         )
 
     def get_backend_config_with_creds(
         self, record: BackendRecord
     ) -> JarvisLabsBackendConfigWithCreds:
         config = self._get_config(record)
-        return JarvisLabsBackendConfigWithCreds.__response__.parse_obj(config)
+        return validate_extra_ignore(JarvisLabsBackendConfigWithCreds, config)
 
     def get_backend_config_without_creds(self, record: BackendRecord) -> JarvisLabsBackendConfig:
         config = self._get_config(record)
-        return JarvisLabsBackendConfig.__response__.parse_obj(config)
+        return validate_extra_ignore(JarvisLabsBackendConfig, config)
 
     def get_backend(self, record: BackendRecord) -> JarvisLabsBackend:
         config = self._get_config(record)
         return JarvisLabsBackend(config=config)
 
     def _get_config(self, record: BackendRecord) -> JarvisLabsConfig:
-        return JarvisLabsConfig.__response__(
-            **json.loads(record.config),
-            creds=JarvisLabsCreds.__response__.parse_raw(record.auth),
+        return validate_extra_ignore(
+            JarvisLabsConfig,
+            {
+                **json.loads(record.config),
+                "creds": validate_json_extra_ignore(JarvisLabsCreds, record.auth),
+            },
         )
 
     def _validate_api_key(self, api_key: str):
diff --git a/src/dstack/_internal/core/backends/kubernetes/compute.py b/src/dstack/_internal/core/backends/kubernetes/compute.py
index 3340aeee5e..3fb02ce8f3 100644
--- a/src/dstack/_internal/core/backends/kubernetes/compute.py
+++ b/src/dstack/_internal/core/backends/kubernetes/compute.py
@@ -77,7 +77,7 @@
 from dstack._internal.core.consts import DSTACK_RUNNER_SSH_PORT
 from dstack._internal.core.errors import ComputeError, ProvisioningError, SkipOffer
 from dstack._internal.core.models.backends.base import BackendType
-from dstack._internal.core.models.common import CoreModel
+from dstack._internal.core.models.common import CoreModel, validate_json_extra_ignore
 from dstack._internal.core.models.gateways import (
     GatewayComputeConfiguration,
     GatewayProvisioningData,
@@ -128,7 +128,7 @@ class KubernetesBackendData(CoreModel):
 
     @classmethod
     def load(cls, raw: str) -> Self:
-        return cls.__response__.parse_raw(raw)
+        return validate_json_extra_ignore(cls, raw)
 
 
 class KubernetesCompute(
@@ -349,7 +349,7 @@ def run_job(
             instance_type=instance_offer.instance,
             internal_ip=None,
             ssh_proxy=None,
-            backend_data=backend_data.json(),
+            backend_data=backend_data.model_dump_json(),
         )
 
     def update_provisioning_data(
@@ -825,7 +825,7 @@ def _get_amd_gpu_node_affinity(
 def _offer_modifier(
     resource_requests: ResourceRequests, offer: InstanceOfferWithAvailability
 ) -> InstanceOfferWithAvailability:
-    offer_copy = offer.copy(deep=True)
+    offer_copy = offer.model_copy(deep=True)
     adjust_resources_by_resource_requests(offer_copy.instance.resources, resource_requests)
     return offer_copy
 
diff --git a/src/dstack/_internal/core/backends/kubernetes/configurator.py b/src/dstack/_internal/core/backends/kubernetes/configurator.py
index b8872f0211..8a4e2e17f0 100644
--- a/src/dstack/_internal/core/backends/kubernetes/configurator.py
+++ b/src/dstack/_internal/core/backends/kubernetes/configurator.py
@@ -16,6 +16,7 @@
 )
 from dstack._internal.core.errors import ServerClientError
 from dstack._internal.core.models.backends.base import BackendType
+from dstack._internal.core.models.common import validate_extra_ignore, validate_json_extra_ignore
 from dstack._internal.utils.logging import get_logger
 
 logger = get_logger(__name__)
@@ -49,7 +50,7 @@ def create_backend(
         self, project_name: str, config: KubernetesBackendConfigWithCreds
     ) -> BackendRecord:
         return BackendRecord(
-            config=KubernetesStoredConfig.__response__.parse_obj(config).json(),
+            config=validate_extra_ignore(KubernetesStoredConfig, config).model_dump_json(),
             auth="",
         )
 
@@ -57,17 +58,17 @@ def get_backend_config_with_creds(
         self, record: BackendRecord
     ) -> KubernetesBackendConfigWithCreds:
         config = self._get_config(record)
-        return KubernetesBackendConfigWithCreds.__response__.parse_obj(config)
+        return validate_extra_ignore(KubernetesBackendConfigWithCreds, config)
 
     def get_backend_config_without_creds(self, record: BackendRecord) -> KubernetesBackendConfig:
         config = self._get_config(record)
-        return KubernetesBackendConfig.__response__.parse_obj(config)
+        return validate_extra_ignore(KubernetesBackendConfig, config)
 
     def get_backend(self, record: BackendRecord) -> KubernetesBackend:
         return KubernetesBackend(self._get_config(record))
 
     def _get_config(self, record: BackendRecord) -> KubernetesConfig:
-        return KubernetesConfig.__response__.parse_raw(record.config)
+        return validate_json_extra_ignore(KubernetesConfig, record.config)
 
     def _check_config_contexts(self, config: KubernetesBackendConfig):
         if config.contexts is None:
diff --git a/src/dstack/_internal/core/backends/kubernetes/models.py b/src/dstack/_internal/core/backends/kubernetes/models.py
index eb92982e45..c374bc8e62 100644
--- a/src/dstack/_internal/core/backends/kubernetes/models.py
+++ b/src/dstack/_internal/core/backends/kubernetes/models.py
@@ -1,6 +1,6 @@
 from typing import Annotated, Literal, Optional, Union
 
-from pydantic import Field, root_validator
+from pydantic import Field, model_validator
 
 from dstack._internal.core.backends.base.models import fill_data
 from dstack._internal.core.models.common import CoreModel
@@ -89,7 +89,8 @@ class KubeconfigFileConfig(CoreModel):
         ),
     ] = None
 
-    @root_validator
+    @model_validator(mode="before")
+    @classmethod
     def fill_data(cls, values: dict) -> dict:
         if values.get("filename") == "" and values.get("data") is None:
             raise ValueError("filename or data must be specified")
diff --git a/src/dstack/_internal/core/backends/kubernetes/utils.py b/src/dstack/_internal/core/backends/kubernetes/utils.py
index a40e8abcbe..b6be7d0e7a 100644
--- a/src/dstack/_internal/core/backends/kubernetes/utils.py
+++ b/src/dstack/_internal/core/backends/kubernetes/utils.py
@@ -30,7 +30,7 @@
     KubernetesBackendConfigWithCreds,
     KubernetesProxyJumpConfig,
 )
-from dstack._internal.core.models.common import CoreModel
+from dstack._internal.core.models.common import CoreModel, validate_extra_ignore
 from dstack._internal.utils.logging import get_logger
 
 logger = get_logger(__name__)
@@ -192,7 +192,7 @@ def kubeconfig_data_to_kubeconfig_dict(kubeconfig_data: str) -> dict:
 
 
 def kubeconfig_dict_to_kubeconfig(kubeconfig_dict: dict) -> Kubeconfig:
-    return Kubeconfig.__response__.parse_obj(kubeconfig_dict)
+    return validate_extra_ignore(Kubeconfig, kubeconfig_dict)
 
 
 def call_api_method(
diff --git a/src/dstack/_internal/core/backends/lambdalabs/configurator.py b/src/dstack/_internal/core/backends/lambdalabs/configurator.py
index 7e4d49c717..bcb8ad4f3f 100644
--- a/src/dstack/_internal/core/backends/lambdalabs/configurator.py
+++ b/src/dstack/_internal/core/backends/lambdalabs/configurator.py
@@ -17,6 +17,7 @@
 from dstack._internal.core.models.backends.base import (
     BackendType,
 )
+from dstack._internal.core.models.common import validate_extra_ignore, validate_json_extra_ignore
 
 
 class LambdaConfigurator(
@@ -36,27 +37,30 @@ def create_backend(
     ) -> BackendRecord:
         return BackendRecord(
             config=LambdaStoredConfig(
-                **LambdaBackendConfig.__response__.parse_obj(config).dict()
-            ).json(),
-            auth=LambdaCreds.parse_obj(config.creds).json(),
+                **validate_extra_ignore(LambdaBackendConfig, config).model_dump()
+            ).model_dump_json(),
+            auth=LambdaCreds.model_validate(config.creds).model_dump_json(),
         )
 
     def get_backend_config_with_creds(self, record: BackendRecord) -> LambdaBackendConfigWithCreds:
         config = self._get_config(record)
-        return LambdaBackendConfigWithCreds.__response__.parse_obj(config)
+        return validate_extra_ignore(LambdaBackendConfigWithCreds, config)
 
     def get_backend_config_without_creds(self, record: BackendRecord) -> LambdaBackendConfig:
         config = self._get_config(record)
-        return LambdaBackendConfig.__response__.parse_obj(config)
+        return validate_extra_ignore(LambdaBackendConfig, config)
 
     def get_backend(self, record: BackendRecord) -> LambdaBackend:
         config = self._get_config(record)
         return LambdaBackend(config=config)
 
     def _get_config(self, record: BackendRecord) -> LambdaConfig:
-        return LambdaConfig.__response__(
-            **json.loads(record.config),
-            creds=LambdaCreds.__response__.parse_raw(record.auth),
+        return validate_extra_ignore(
+            LambdaConfig,
+            {
+                **json.loads(record.config),
+                "creds": validate_json_extra_ignore(LambdaCreds, record.auth),
+            },
         )
 
     def _validate_lambda_api_key(self, api_key: str):
diff --git a/src/dstack/_internal/core/backends/models.py b/src/dstack/_internal/core/backends/models.py
index bf06dbf3f0..25824a1eea 100644
--- a/src/dstack/_internal/core/backends/models.py
+++ b/src/dstack/_internal/core/backends/models.py
@@ -1,6 +1,6 @@
 from typing import Annotated, Union
 
-from pydantic import Field
+from pydantic import Field, RootModel
 
 from dstack._internal.core.backends.aws.models import (
     AWSBackendConfig,
@@ -142,43 +142,23 @@
     DstackBackendConfig,
 ]
 
-# Permissive counterpart of `AnyBackendConfigWithCreds` for parsing server responses.
-# A newer server may add config fields that an older client's models don't know about;
-# parsing with the strict variant would reject the response outright.
+# The same union tagged for validation. Without the discriminator, arm selection would depend on
+# trying each of the 20 arms in order and reporting 20 errors when none match.
 #
-# Discriminated on `type`: without it, arm selection would depend on trying each of the 20
-# arms in order, which only works because every arm happens to declare a `Literal` type.
-# `AnyBackendConfigWithCreds` above stays a bare `Union` on purpose. Its two server-side users apply
-# `Field(discriminator="type")` at the point of use, which is fine against a bare alias.
-# Baking the discriminator into the alias would turn those into doubled `Annotated` `Field`s and
-# fail with `ValueError: cannot specify multiple 'Annotated' 'Field's`.
-# Discriminating here is because nothing else wraps this alias.
-AnyBackendConfigWithCredsResponse = Annotated[
-    Union[
-        AWSBackendConfigWithCreds.__response__,
-        AzureBackendConfigWithCreds.__response__,
-        CloudRiftBackendConfigWithCreds.__response__,
-        CrusoeBackendConfigWithCreds.__response__,
-        CudoBackendConfigWithCreds.__response__,
-        VerdaBackendConfigWithCreds.__response__,
-        BaseDigitalOceanBackendConfigWithCreds.__response__,
-        GCPBackendConfigWithCreds.__response__,
-        HotAisleBackendConfigWithCreds.__response__,
-        JarvisLabsBackendConfigWithCreds.__response__,
-        KubernetesBackendConfigWithCreds.__response__,
-        LambdaBackendConfigWithCreds.__response__,
-        OCIBackendConfigWithCreds.__response__,
-        NebiusBackendConfigWithCreds.__response__,
-        RunpodBackendConfigWithCreds.__response__,
-        TensorDockBackendConfigWithCreds.__response__,
-        VastAIBackendConfigWithCreds.__response__,
-        VultrBackendConfigWithCreds.__response__,
-        SlurmBackendConfigWithCreds.__response__,
-        DstackBackendConfig.__response__,
-    ],
+# `AnyBackendConfigWithCreds` above stays a bare `Union` because it is also used as a plain type
+# annotation and as the bound of `BackendConfigWithCredsT` in `base/configurator.py`. Every site
+# that *validates* the union should use this alias instead of wrapping it again locally: two
+# `Annotated` `Field`s on the same type fail with "cannot specify multiple 'Annotated' 'Field's".
+AnyBackendConfigWithCredsTagged = Annotated[
+    AnyBackendConfigWithCreds,
     Field(discriminator="type"),
 ]
 
+
+class BackendConfigWithCreds(RootModel[AnyBackendConfigWithCredsTagged]):
+    pass
+
+
 # Backend config accepted in server/config.yaml.
 # This can be different from the API config.
 # For example, it can make creds data optional and resolve it by filename.
diff --git a/src/dstack/_internal/core/backends/nebius/compute.py b/src/dstack/_internal/core/backends/nebius/compute.py
index a12eb7a205..620884945b 100644
--- a/src/dstack/_internal/core/backends/nebius/compute.py
+++ b/src/dstack/_internal/core/backends/nebius/compute.py
@@ -40,7 +40,11 @@
     ProvisioningError,
 )
 from dstack._internal.core.models.backends.base import BackendType
-from dstack._internal.core.models.common import CoreModel
+from dstack._internal.core.models.common import (
+    CoreModel,
+    validate_extra_ignore,
+    validate_json_extra_ignore,
+)
 from dstack._internal.core.models.instances import (
     InstanceAvailability,
     InstanceConfiguration,
@@ -240,7 +244,9 @@ def create_instance(
             ssh_port=22,
             username="ubuntu",
             dockerized=True,
-            backend_data=NebiusInstanceBackendData(boot_disk_id=create_disk_op.resource_id).json(),
+            backend_data=NebiusInstanceBackendData(
+                boot_disk_id=create_disk_op.resource_id
+            ).model_dump_json(),
         )
 
     def update_provisioning_data(
@@ -285,8 +291,8 @@ def create_placement_group(
         master_instance_offer: InstanceOffer,
     ) -> PlacementGroupProvisioningData:
         assert placement_group.configuration.placement_strategy == PlacementStrategy.CLUSTER
-        master_instance_offer_backend_data: NebiusOfferBackendData = (
-            NebiusOfferBackendData.__response__.parse_obj(master_instance_offer.backend_data)
+        master_instance_offer_backend_data: NebiusOfferBackendData = validate_extra_ignore(
+            NebiusOfferBackendData, master_instance_offer.backend_data
         )
         fabrics = list(master_instance_offer_backend_data.fabrics)
         if self.config.fabrics is not None:
@@ -308,7 +314,7 @@ def create_placement_group(
             )
         return PlacementGroupProvisioningData(
             backend=BackendType.NEBIUS,
-            backend_data=placement_group_backend_data.json(),
+            backend_data=placement_group_backend_data.model_dump_json(),
         )
 
     def delete_placement_group(self, placement_group: PlacementGroup) -> None:
@@ -331,8 +337,8 @@ def is_suitable_placement_group(
         placement_group_backend_data = NebiusPlacementGroupBackendData.load(
             placement_group.provisioning_data.backend_data
         )
-        instance_offer_backend_data: NebiusOfferBackendData = (
-            NebiusOfferBackendData.__response__.parse_obj(instance_offer.backend_data)
+        instance_offer_backend_data: NebiusOfferBackendData = validate_extra_ignore(
+            NebiusOfferBackendData, instance_offer.backend_data
         )
         return (
             placement_group_backend_data.cluster is None
@@ -346,7 +352,7 @@ class NebiusInstanceBackendData(CoreModel):
     @classmethod
     def load(cls, raw: Optional[str]) -> "NebiusInstanceBackendData":
         assert raw is not None
-        return cls.__response__.parse_raw(raw)
+        return validate_json_extra_ignore(cls, raw)
 
 
 class NebiusClusterBackendData(CoreModel):
@@ -360,7 +366,7 @@ class NebiusPlacementGroupBackendData(CoreModel):
     @classmethod
     def load(cls, raw: Optional[str]) -> "NebiusPlacementGroupBackendData":
         assert raw is not None
-        return cls.__response__.parse_raw(raw)
+        return validate_json_extra_ignore(cls, raw)
 
 
 def _wait_for_instance(sdk: SDK, op: SDKOperation[Operation]) -> None:
diff --git a/src/dstack/_internal/core/backends/nebius/configurator.py b/src/dstack/_internal/core/backends/nebius/configurator.py
index a349863527..483d4c4137 100644
--- a/src/dstack/_internal/core/backends/nebius/configurator.py
+++ b/src/dstack/_internal/core/backends/nebius/configurator.py
@@ -21,6 +21,7 @@
 from dstack._internal.core.backends.nebius.resources import get_all_infiniband_fabrics
 from dstack._internal.core.errors import BackendError, ServerClientError
 from dstack._internal.core.models.backends.base import BackendType
+from dstack._internal.core.models.common import validate_extra_ignore, validate_json_extra_ignore
 
 
 class NebiusConfigurator(
@@ -74,25 +75,28 @@ def create_backend(
     ) -> BackendRecord:
         return BackendRecord(
             config=NebiusStoredConfig(
-                **NebiusBackendConfig.__response__.parse_obj(config).dict()
-            ).json(),
-            auth=NebiusCreds.parse_obj(config.creds).json(),
+                **validate_extra_ignore(NebiusBackendConfig, config).model_dump()
+            ).model_dump_json(),
+            auth=NebiusCreds.model_validate(config.creds).model_dump_json(),
         )
 
     def get_backend_config_with_creds(self, record: BackendRecord) -> NebiusBackendConfigWithCreds:
         config = self._get_config(record)
-        return NebiusBackendConfigWithCreds.__response__.parse_obj(config)
+        return validate_extra_ignore(NebiusBackendConfigWithCreds, config)
 
     def get_backend_config_without_creds(self, record: BackendRecord) -> NebiusBackendConfig:
         config = self._get_config(record)
-        return NebiusBackendConfig.__response__.parse_obj(config)
+        return validate_extra_ignore(NebiusBackendConfig, config)
 
     def get_backend(self, record: BackendRecord) -> NebiusBackend:
         config = self._get_config(record)
         return NebiusBackend(config=config)
 
     def _get_config(self, record: BackendRecord) -> NebiusConfig:
-        return NebiusConfig.__response__(
-            **json.loads(record.config),
-            creds=NebiusCreds.__response__.parse_raw(record.auth),
+        return validate_extra_ignore(
+            NebiusConfig,
+            {
+                **json.loads(record.config),
+                "creds": validate_json_extra_ignore(NebiusCreds, record.auth),
+            },
         )
diff --git a/src/dstack/_internal/core/backends/nebius/models.py b/src/dstack/_internal/core/backends/nebius/models.py
index 143eb55746..9adbc12a05 100644
--- a/src/dstack/_internal/core/backends/nebius/models.py
+++ b/src/dstack/_internal/core/backends/nebius/models.py
@@ -2,7 +2,7 @@
 from pathlib import Path
 from typing import Annotated, Dict, Literal, Optional, Union
 
-from pydantic import Field, root_validator
+from pydantic import Field, field_serializer, model_validator
 
 from dstack._internal.core.backends.base.models import fill_data
 from dstack._internal.core.models.common import CoreModel
@@ -76,7 +76,8 @@ class NebiusServiceAccountFileCreds(CoreModel):
         Optional[str], Field(description="The path to the service account credentials file")
     ] = None
 
-    @root_validator
+    @model_validator(mode="before")
+    @classmethod
     def fill_data(cls, values):
         if filename := values.get("filename"):
             try:
@@ -183,3 +184,7 @@ class NebiusConfig(NebiusStoredConfig):
 
 class NebiusOfferBackendData(CoreModel):
     fabrics: set[str] = set()
+
+    @field_serializer("fabrics")
+    def _serialize_fabrics(self, value: set[str]) -> list[str]:
+        return sorted(value)
diff --git a/src/dstack/_internal/core/backends/nebius/resources.py b/src/dstack/_internal/core/backends/nebius/resources.py
index c9871f2da8..9450f5b45d 100644
--- a/src/dstack/_internal/core/backends/nebius/resources.py
+++ b/src/dstack/_internal/core/backends/nebius/resources.py
@@ -57,6 +57,7 @@
 )
 from dstack._internal.core.errors import BackendError, NoCapacityError
 from dstack._internal.core.models.backends.base import BackendType
+from dstack._internal.core.models.common import validate_extra_ignore
 from dstack._internal.utils.event_loop import DaemonEventLoop
 from dstack._internal.utils.logging import get_logger
 
@@ -257,8 +258,8 @@ def get_all_infiniband_fabrics() -> set[str]:
     offers = get_catalog_offers(backend=BackendType.NEBIUS)
     result = set()
     for offer in offers:
-        backend_data: NebiusOfferBackendData = NebiusOfferBackendData.__response__.parse_obj(
-            offer.backend_data
+        backend_data: NebiusOfferBackendData = validate_extra_ignore(
+            NebiusOfferBackendData, offer.backend_data
         )
         result |= backend_data.fabrics
     return result
diff --git a/src/dstack/_internal/core/backends/oci/auth.py b/src/dstack/_internal/core/backends/oci/auth.py
index c751c10a35..9681ef8eec 100644
--- a/src/dstack/_internal/core/backends/oci/auth.py
+++ b/src/dstack/_internal/core/backends/oci/auth.py
@@ -8,7 +8,7 @@
 def get_client_config(creds: AnyOCICreds) -> Mapping[str, Any]:
     if isinstance(creds, OCIDefaultCreds):
         return oci.config.from_file(file_location=creds.file, profile_name=creds.profile)
-    return creds.dict(exclude={"type"})
+    return creds.model_dump(exclude={"type"})
 
 
 def creds_valid(creds: AnyOCICreds) -> bool:
diff --git a/src/dstack/_internal/core/backends/oci/configurator.py b/src/dstack/_internal/core/backends/oci/configurator.py
index 61ee596ad6..95696330e7 100644
--- a/src/dstack/_internal/core/backends/oci/configurator.py
+++ b/src/dstack/_internal/core/backends/oci/configurator.py
@@ -26,6 +26,7 @@
 from dstack._internal.core.models.backends.base import (
     BackendType,
 )
+from dstack._internal.core.models.common import validate_extra_ignore, validate_json_extra_ignore
 
 # where dstack images are published
 SUPPORTED_REGIONS = frozenset(
@@ -78,31 +79,38 @@ def create_backend(
             project_name, config, subscribed_regions.home_region_name
         )
         config.compartment_id = compartment_id
-        stored_config = OCIStoredConfig.__response__(
-            **config.dict(), subnet_ids_per_region=subnet_ids_per_region
+        stored_config = validate_extra_ignore(
+            OCIStoredConfig,
+            {
+                **config.model_dump(),
+                "subnet_ids_per_region": subnet_ids_per_region,
+            },
         )
 
         return BackendRecord(
-            config=stored_config.json(),
-            auth=OCICreds.parse_obj(config.creds).json(),
+            config=stored_config.model_dump_json(),
+            auth=OCICreds.model_validate(config.creds).model_dump_json(),
         )
 
     def get_backend_config_with_creds(self, record: BackendRecord) -> OCIBackendConfigWithCreds:
         config = self._get_config(record)
-        return OCIBackendConfigWithCreds.__response__.parse_obj(config)
+        return validate_extra_ignore(OCIBackendConfigWithCreds, config)
 
     def get_backend_config_without_creds(self, record: BackendRecord) -> OCIBackendConfig:
         config = self._get_config(record)
-        return OCIBackendConfig.__response__.parse_obj(config)
+        return validate_extra_ignore(OCIBackendConfig, config)
 
     def get_backend(self, record: BackendRecord) -> OCIBackend:
         config = self._get_config(record)
         return OCIBackend(config=config)
 
     def _get_config(self, record: BackendRecord) -> OCIConfig:
-        return OCIConfig.__response__(
-            **json.loads(record.config),
-            creds=OCICreds.__response__.parse_raw(record.auth).__root__,
+        return validate_extra_ignore(
+            OCIConfig,
+            {
+                **json.loads(record.config),
+                "creds": validate_json_extra_ignore(OCICreds, record.auth).root,
+            },
         )
 
 
diff --git a/src/dstack/_internal/core/backends/oci/models.py b/src/dstack/_internal/core/backends/oci/models.py
index 4212efecb3..20f22b793d 100644
--- a/src/dstack/_internal/core/backends/oci/models.py
+++ b/src/dstack/_internal/core/backends/oci/models.py
@@ -1,6 +1,7 @@
 from typing import Annotated, Dict, List, Literal, Optional, Union
 
-from pydantic import Field, root_validator
+from pydantic import Field, RootModel, model_validator
+from typing_extensions import Self
 
 from dstack._internal.core.models.common import CoreModel
 
@@ -29,14 +30,14 @@ class OCIClientCreds(CoreModel):
         str, Field(description="Name or key of any region the tenancy is subscribed to")
     ]
 
-    @root_validator
-    def key_file_xor_key_content(cls, values):
-        key_file, key_content = values["key_file"], values["key_content"]
+    @model_validator(mode="after")
+    def key_file_xor_key_content(self) -> Self:
+        key_file, key_content = self.key_file, self.key_content
         if key_file and key_content:
             raise ValueError("key_file and key_content are mutually exclusive")
         if not key_file and not key_content:
             raise ValueError("Either key_file or key_content should be set")
-        return values
+        return self
 
 
 class OCIDefaultCreds(CoreModel):
@@ -50,8 +51,8 @@ class OCIDefaultCreds(CoreModel):
 AnyOCICreds = Union[OCIClientCreds, OCIDefaultCreds]
 
 
-class OCICreds(CoreModel):
-    __root__: AnyOCICreds = Field(..., discriminator="type")
+class OCICreds(RootModel[Annotated[AnyOCICreds, Field(discriminator="type")]]):
+    pass
 
 
 class OCIBackendConfig(CoreModel):
diff --git a/src/dstack/_internal/core/backends/runpod/compute.py b/src/dstack/_internal/core/backends/runpod/compute.py
index b918139e1b..e0ccc2a41c 100644
--- a/src/dstack/_internal/core/backends/runpod/compute.py
+++ b/src/dstack/_internal/core/backends/runpod/compute.py
@@ -28,7 +28,7 @@
     ComputeError,
 )
 from dstack._internal.core.models.backends.base import BackendType
-from dstack._internal.core.models.common import CoreModel, RegistryAuth
+from dstack._internal.core.models.common import CoreModel, RegistryAuth, validate_extra_ignore
 from dstack._internal.core.models.compute_groups import ComputeGroup, ComputeGroupProvisioningData
 from dstack._internal.core.models.instances import (
     InstanceAvailability,
@@ -485,8 +485,8 @@ def _is_secure_cloud(region: str) -> bool:
 
 
 def _get_offer_pod_counts(offer: InstanceOfferWithAvailability) -> list[int]:
-    backend_data: RunpodOfferBackendData = RunpodOfferBackendData.__response__.parse_obj(
-        offer.backend_data
+    backend_data: RunpodOfferBackendData = validate_extra_ignore(
+        RunpodOfferBackendData, offer.backend_data
     )
     pod_counts = backend_data.pod_counts or []
     return pod_counts
diff --git a/src/dstack/_internal/core/backends/runpod/configurator.py b/src/dstack/_internal/core/backends/runpod/configurator.py
index c2c3221318..eb4dfd7dba 100644
--- a/src/dstack/_internal/core/backends/runpod/configurator.py
+++ b/src/dstack/_internal/core/backends/runpod/configurator.py
@@ -15,6 +15,7 @@
     RunpodStoredConfig,
 )
 from dstack._internal.core.models.backends.base import BackendType
+from dstack._internal.core.models.common import validate_extra_ignore, validate_json_extra_ignore
 
 
 class RunpodConfigurator(
@@ -34,27 +35,30 @@ def create_backend(
     ) -> BackendRecord:
         return BackendRecord(
             config=RunpodStoredConfig(
-                **RunpodBackendConfig.__response__.parse_obj(config).dict()
-            ).json(),
-            auth=RunpodCreds.parse_obj(config.creds).json(),
+                **validate_extra_ignore(RunpodBackendConfig, config).model_dump()
+            ).model_dump_json(),
+            auth=RunpodCreds.model_validate(config.creds).model_dump_json(),
         )
 
     def get_backend_config_with_creds(self, record: BackendRecord) -> RunpodBackendConfigWithCreds:
         config = self._get_config(record)
-        return RunpodBackendConfigWithCreds.__response__.parse_obj(config)
+        return validate_extra_ignore(RunpodBackendConfigWithCreds, config)
 
     def get_backend_config_without_creds(self, record: BackendRecord) -> RunpodBackendConfig:
         config = self._get_config(record)
-        return RunpodBackendConfig.__response__.parse_obj(config)
+        return validate_extra_ignore(RunpodBackendConfig, config)
 
     def get_backend(self, record: BackendRecord) -> RunpodBackend:
         config = self._get_config(record)
         return RunpodBackend(config=config)
 
     def _get_config(self, record: BackendRecord) -> RunpodConfig:
-        return RunpodConfig.__response__(
-            **json.loads(record.config),
-            creds=RunpodCreds.__response__.parse_raw(record.auth),
+        return validate_extra_ignore(
+            RunpodConfig,
+            {
+                **json.loads(record.config),
+                "creds": validate_json_extra_ignore(RunpodCreds, record.auth),
+            },
         )
 
     def _validate_runpod_api_key(self, api_key: str):
diff --git a/src/dstack/_internal/core/backends/slurm/compute.py b/src/dstack/_internal/core/backends/slurm/compute.py
index ca9088b8a7..3ae14d5d99 100644
--- a/src/dstack/_internal/core/backends/slurm/compute.py
+++ b/src/dstack/_internal/core/backends/slurm/compute.py
@@ -361,7 +361,7 @@ def _offer_modifier(
         if not filtered_partitions:
             return None
 
-        offer_copy = offer.copy(deep=True)
+        offer_copy = offer.model_copy(deep=True)
         _adjust_resources(offer_copy.instance.resources, requested_resources)
         offer_copy.availability_zones = list(filtered_partitions)
         return offer_copy
diff --git a/src/dstack/_internal/core/backends/slurm/configurator.py b/src/dstack/_internal/core/backends/slurm/configurator.py
index 084dcf5552..99bd7f957b 100644
--- a/src/dstack/_internal/core/backends/slurm/configurator.py
+++ b/src/dstack/_internal/core/backends/slurm/configurator.py
@@ -14,6 +14,7 @@
 )
 from dstack._internal.core.errors import ServerClientError
 from dstack._internal.core.models.backends.base import BackendType
+from dstack._internal.core.models.common import validate_extra_ignore, validate_json_extra_ignore
 
 
 class SlurmConfigurator(
@@ -36,23 +37,23 @@ def create_backend(
         self, project_name: str, config: SlurmBackendConfigWithCreds
     ) -> BackendRecord:
         return BackendRecord(
-            config=SlurmStoredConfig.__response__.parse_obj(config).json(),
+            config=validate_extra_ignore(SlurmStoredConfig, config).model_dump_json(),
             auth="",
         )
 
     def get_backend_config_with_creds(self, record: BackendRecord) -> SlurmBackendConfigWithCreds:
         config = self._get_config(record)
-        return SlurmBackendConfigWithCreds.__response__.parse_obj(config)
+        return validate_extra_ignore(SlurmBackendConfigWithCreds, config)
 
     def get_backend_config_without_creds(self, record: BackendRecord) -> SlurmBackendConfig:
         config = self._get_config(record)
-        return SlurmBackendConfig.__response__.parse_obj(config)
+        return validate_extra_ignore(SlurmBackendConfig, config)
 
     def get_backend(self, record: BackendRecord) -> SlurmBackend:
         return SlurmBackend(self._get_config(record))
 
     def _get_config(self, record: BackendRecord) -> SlurmConfig:
-        return SlurmConfig.__response__.parse_raw(record.config)
+        return validate_json_extra_ignore(SlurmConfig, record.config)
 
     def _check_clusters(self, clusters: list[SlurmCluster]) -> None:
         error_messages: list[str] = []
diff --git a/src/dstack/_internal/core/backends/slurm/models.py b/src/dstack/_internal/core/backends/slurm/models.py
index 485fd96bcd..f34e6b7e92 100644
--- a/src/dstack/_internal/core/backends/slurm/models.py
+++ b/src/dstack/_internal/core/backends/slurm/models.py
@@ -1,6 +1,6 @@
 from typing import Annotated, Literal, Optional, Union
 
-from pydantic import Field, root_validator
+from pydantic import Field, model_validator
 
 from dstack._internal.core.backends.base.models import fill_data
 from dstack._internal.core.models.common import CoreModel
@@ -104,7 +104,8 @@ class SlurmPrivateKeyFileConfig(CoreModel):
         ),
     ] = None
 
-    @root_validator
+    @model_validator(mode="before")
+    @classmethod
     def fill_data(cls, values: dict) -> dict:
         return fill_data(values, filename_field="path", data_field="content")
 
diff --git a/src/dstack/_internal/core/backends/template/configurator.py.jinja b/src/dstack/_internal/core/backends/template/configurator.py.jinja
index 47ea303903..8206004261 100644
--- a/src/dstack/_internal/core/backends/template/configurator.py.jinja
+++ b/src/dstack/_internal/core/backends/template/configurator.py.jinja
@@ -17,6 +17,7 @@ from dstack._internal.core.backends.{{ backend_name|lower }}.models import (
 from dstack._internal.core.models.backends.base import (
     BackendType,
 )
+from dstack._internal.core.models.common import validate_extra_ignore, validate_json_extra_ignore
 
 
 class {{ backend_name }}Configurator(
@@ -39,27 +40,30 @@ class {{ backend_name }}Configurator(
     ) -> BackendRecord:
         return BackendRecord(
             config={{ backend_name }}StoredConfig(
-                **{{ backend_name }}BackendConfig.__response__.parse_obj(config).dict()
-            ).json(),
-            auth={{ backend_name }}Creds.parse_obj(config.creds).json(),
+                **validate_extra_ignore({{ backend_name }}BackendConfig, config).model_dump()
+            ).model_dump_json(),
+            auth={{ backend_name }}Creds.model_validate(config.creds).model_dump_json(),
         )
 
     def get_backend_config_with_creds(self, record: BackendRecord) -> {{ backend_name }}BackendConfigWithCreds:
         config = self._get_config(record)
-        return {{ backend_name }}BackendConfigWithCreds.__response__.parse_obj(config)
+        return validate_extra_ignore({{ backend_name }}BackendConfigWithCreds, config)
 
     def get_backend_config_without_creds(self, record: BackendRecord) -> {{ backend_name }}BackendConfig:
         config = self._get_config(record)
-        return {{ backend_name }}BackendConfig.__response__.parse_obj(config)
+        return validate_extra_ignore({{ backend_name }}BackendConfig, config)
 
     def get_backend(self, record: BackendRecord) -> {{ backend_name }}Backend:
         config = self._get_config(record)
         return {{ backend_name }}Backend(config=config)
 
     def _get_config(self, record: BackendRecord) -> {{ backend_name }}Config:
-        return {{ backend_name }}Config.__response__(
-            **json.loads(record.config),
-            creds={{ backend_name }}Creds.parse_raw(record.auth),
+        return validate_extra_ignore(
+            {{ backend_name }}Config,
+            {
+                **json.loads(record.config),
+                "creds": validate_json_extra_ignore({{ backend_name }}Creds, record.auth),
+            },
         )
 
     def _validate_creds(self, creds: Any{{ backend_name }}Creds):
diff --git a/src/dstack/_internal/core/backends/vastai/compute.py b/src/dstack/_internal/core/backends/vastai/compute.py
index bf85fc315a..a419b84bec 100644
--- a/src/dstack/_internal/core/backends/vastai/compute.py
+++ b/src/dstack/_internal/core/backends/vastai/compute.py
@@ -28,7 +28,7 @@
 from dstack._internal.core.consts import DSTACK_RUNNER_SSH_PORT
 from dstack._internal.core.errors import ComputeError, NoCapacityError, ProvisioningError
 from dstack._internal.core.models.backends.base import BackendType
-from dstack._internal.core.models.common import CoreModel
+from dstack._internal.core.models.common import CoreModel, validate_extra_ignore
 from dstack._internal.core.models.instances import (
     InstanceAvailability,
     InstanceOfferWithAvailability,
@@ -131,8 +131,8 @@ def run_job(
         commands = get_docker_commands(
             [run.run_spec.ssh_key_pub.strip(), project_ssh_public_key.strip()]
         )
-        offer_backend_data: VastAIOfferBackendData = VastAIOfferBackendData.__response__.parse_obj(
-            instance_offer.backend_data
+        offer_backend_data: VastAIOfferBackendData = validate_extra_ignore(
+            VastAIOfferBackendData, instance_offer.backend_data
         )
         bid = None
         if instance_offer.instance.resources.spot:
diff --git a/src/dstack/_internal/core/backends/vastai/configurator.py b/src/dstack/_internal/core/backends/vastai/configurator.py
index cab9dbb10a..6dde3c0894 100644
--- a/src/dstack/_internal/core/backends/vastai/configurator.py
+++ b/src/dstack/_internal/core/backends/vastai/configurator.py
@@ -17,6 +17,7 @@
 from dstack._internal.core.models.backends.base import (
     BackendType,
 )
+from dstack._internal.core.models.common import validate_extra_ignore, validate_json_extra_ignore
 
 REGIONS = []
 
@@ -40,27 +41,30 @@ def create_backend(
             config.regions = REGIONS
         return BackendRecord(
             config=VastAIStoredConfig(
-                **VastAIBackendConfig.__response__.parse_obj(config).dict()
-            ).json(),
-            auth=VastAICreds.parse_obj(config.creds).json(),
+                **validate_extra_ignore(VastAIBackendConfig, config).model_dump()
+            ).model_dump_json(),
+            auth=VastAICreds.model_validate(config.creds).model_dump_json(),
         )
 
     def get_backend_config_with_creds(self, record: BackendRecord) -> VastAIBackendConfigWithCreds:
         config = self._get_config(record)
-        return VastAIBackendConfigWithCreds.__response__.parse_obj(config)
+        return validate_extra_ignore(VastAIBackendConfigWithCreds, config)
 
     def get_backend_config_without_creds(self, record: BackendRecord) -> VastAIBackendConfig:
         config = self._get_config(record)
-        return VastAIBackendConfig.__response__.parse_obj(config)
+        return validate_extra_ignore(VastAIBackendConfig, config)
 
     def get_backend(self, record: BackendRecord) -> VastAIBackend:
         config = self._get_config(record)
         return VastAIBackend(config=config)
 
     def _get_config(self, record: BackendRecord) -> VastAIConfig:
-        return VastAIConfig.__response__(
-            **json.loads(record.config),
-            creds=VastAICreds.__response__.parse_raw(record.auth),
+        return validate_extra_ignore(
+            VastAIConfig,
+            {
+                **json.loads(record.config),
+                "creds": validate_json_extra_ignore(VastAICreds, record.auth),
+            },
         )
 
     def _validate_vastai_creds(self, api_key: str):
diff --git a/src/dstack/_internal/core/backends/verda/compute.py b/src/dstack/_internal/core/backends/verda/compute.py
index f6cb4c19ac..c8b71b4058 100644
--- a/src/dstack/_internal/core/backends/verda/compute.py
+++ b/src/dstack/_internal/core/backends/verda/compute.py
@@ -27,7 +27,7 @@
     ProvisioningError,
 )
 from dstack._internal.core.models.backends.base import BackendType
-from dstack._internal.core.models.common import CoreModel
+from dstack._internal.core.models.common import CoreModel, validate_json_extra_ignore
 from dstack._internal.core.models.instances import (
     InstanceAvailability,
     InstanceConfiguration,
@@ -193,7 +193,7 @@ def create_instance(
             backend_data=VerdaInstanceBackendData(
                 startup_script_id=startup_script_id,
                 ssh_key_ids=ssh_ids,
-            ).json(),
+            ).model_dump_json(),
         )
 
     def terminate_instance(
@@ -330,4 +330,4 @@ class VerdaInstanceBackendData(CoreModel):
     def load(cls, raw: Optional[str]) -> "VerdaInstanceBackendData":
         if raw is None:
             return cls()
-        return cls.__response__.parse_raw(raw)
+        return validate_json_extra_ignore(cls, raw)
diff --git a/src/dstack/_internal/core/backends/verda/configurator.py b/src/dstack/_internal/core/backends/verda/configurator.py
index 274c96638d..f427f43691 100644
--- a/src/dstack/_internal/core/backends/verda/configurator.py
+++ b/src/dstack/_internal/core/backends/verda/configurator.py
@@ -19,6 +19,7 @@
 from dstack._internal.core.models.backends.base import (
     BackendType,
 )
+from dstack._internal.core.models.common import validate_extra_ignore, validate_json_extra_ignore
 
 
 class VerdaConfigurator(
@@ -38,27 +39,30 @@ def create_backend(
     ) -> BackendRecord:
         return BackendRecord(
             config=VerdaStoredConfig(
-                **VerdaBackendConfig.__response__.parse_obj(config).dict()
-            ).json(),
-            auth=VerdaCreds.parse_obj(config.creds).json(),
+                **validate_extra_ignore(VerdaBackendConfig, config).model_dump()
+            ).model_dump_json(),
+            auth=VerdaCreds.model_validate(config.creds).model_dump_json(),
         )
 
     def get_backend_config_with_creds(self, record: BackendRecord) -> VerdaBackendConfigWithCreds:
         config = self._get_config(record)
-        return VerdaBackendConfigWithCreds.__response__.parse_obj(config)
+        return validate_extra_ignore(VerdaBackendConfigWithCreds, config)
 
     def get_backend_config_without_creds(self, record: BackendRecord) -> VerdaBackendConfig:
         config = self._get_config(record)
-        return VerdaBackendConfig.__response__.parse_obj(config)
+        return validate_extra_ignore(VerdaBackendConfig, config)
 
     def get_backend(self, record: BackendRecord) -> VerdaBackend:
         config = self._get_config(record)
         return VerdaBackend(config=config)
 
     def _get_config(self, record: BackendRecord) -> VerdaConfig:
-        return VerdaConfig.__response__(
-            **json.loads(record.config),
-            creds=VerdaCreds.__response__.parse_raw(record.auth),
+        return validate_extra_ignore(
+            VerdaConfig,
+            {
+                **json.loads(record.config),
+                "creds": validate_json_extra_ignore(VerdaCreds, record.auth),
+            },
         )
 
     def _validate_creds(self, creds: VerdaCreds):
diff --git a/src/dstack/_internal/core/backends/vultr/configurator.py b/src/dstack/_internal/core/backends/vultr/configurator.py
index 23bde1c381..9d70b2886f 100644
--- a/src/dstack/_internal/core/backends/vultr/configurator.py
+++ b/src/dstack/_internal/core/backends/vultr/configurator.py
@@ -19,6 +19,7 @@
 from dstack._internal.core.models.backends.base import (
     BackendType,
 )
+from dstack._internal.core.models.common import validate_extra_ignore, validate_json_extra_ignore
 
 REGIONS = []
 
@@ -42,27 +43,30 @@ def create_backend(
             config.regions = REGIONS
         return BackendRecord(
             config=VultrStoredConfig(
-                **VultrBackendConfig.__response__.parse_obj(config).dict()
-            ).json(),
-            auth=VultrCreds.parse_obj(config.creds).json(),
+                **validate_extra_ignore(VultrBackendConfig, config).model_dump()
+            ).model_dump_json(),
+            auth=VultrCreds.model_validate(config.creds).model_dump_json(),
         )
 
     def get_backend_config_with_creds(self, record: BackendRecord) -> VultrBackendConfigWithCreds:
         config = self._get_config(record)
-        return VultrBackendConfigWithCreds.__response__.parse_obj(config)
+        return validate_extra_ignore(VultrBackendConfigWithCreds, config)
 
     def get_backend_config_without_creds(self, record: BackendRecord) -> VultrBackendConfig:
         config = self._get_config(record)
-        return VultrBackendConfig.__response__.parse_obj(config)
+        return validate_extra_ignore(VultrBackendConfig, config)
 
     def get_backend(self, record: BackendRecord) -> VultrBackend:
         config = self._get_config(record)
         return VultrBackend(config=config)
 
     def _get_config(self, record: BackendRecord) -> VultrConfig:
-        return VultrConfig.__response__(
-            **json.loads(record.config),
-            creds=VultrCreds.__response__.parse_raw(record.auth),
+        return validate_extra_ignore(
+            VultrConfig,
+            {
+                **json.loads(record.config),
+                "creds": validate_json_extra_ignore(VultrCreds, record.auth),
+            },
         )
 
     def _validate_vultr_api_key(self, api_key: str):
diff --git a/src/dstack/_internal/core/models/common.py b/src/dstack/_internal/core/models/common.py
index 42c56765d2..69e4297dca 100644
--- a/src/dstack/_internal/core/models/common.py
+++ b/src/dstack/_internal/core/models/common.py
@@ -1,123 +1,118 @@
-import re
 from enum import Enum
-from typing import TYPE_CHECKING, Any, Callable, Mapping, Optional, Union
-
-import orjson
-from pydantic import Field
-from pydantic_duality import generate_dual_base_model
+from typing import Any, Optional, TypeVar, Union, overload
+
+from pydantic import (
+    BaseModel,
+    ConfigDict,
+    Field,
+    TypeAdapter,
+)
 from typing_extensions import Annotated
 
-from dstack._internal.utils.json_utils import pydantic_orjson_dumps
+# pydantic v2 generates draft 2020-12. The published `configuration.json` / `profiles.json`
+# advertise the dialect they were generated for, so this has to move with pydantic.
+JSON_SCHEMA_DIALECT = "https://json-schema.org/draft/2020-12/schema"
+
+
+def drop_merged_profile(schema: dict[str, Any]) -> None:
+    """
+    `json_schema_extra` hook for the specs carrying a `merged_profile`.
+
+    It is an internal field computed from the configuration and the profile, never written by a
+    user, so it must not appear in the published schema.
+    """
+    schema.get("properties", {}).pop("merged_profile", None)
+
 
+# Mirrors pydantic v2's `IncEx`, so these can be passed straight to `model_dump`/`model_copy`.
+# v2 keys a mapping by int or str but not both, and its values are `IncEx | bool`.
 IncludeExcludeFieldType = Union[int, str]
-IncludeExcludeSetType = set[IncludeExcludeFieldType]
-IncludeExcludeDictType = dict[
-    IncludeExcludeFieldType, Union[bool, IncludeExcludeSetType, "IncludeExcludeDictType"]
+IncludeExcludeSetType = Union[set[int], set[str]]
+# `dict` rather than `Mapping`, so these stay assignable both *to* pydantic's `IncEx` and to the
+# plain `Dict` parameters the plugin API declares. Keyed by int or str but not both, like `IncEx`.
+IncludeExcludeDictType = Union[
+    dict[int, Union["IncludeExcludeType", bool]],
+    dict[str, Union["IncludeExcludeType", bool]],
 ]
 IncludeExcludeType = Union[IncludeExcludeSetType, IncludeExcludeDictType]
 
 
-class CoreConfig:
-    json_loads = orjson.loads
-    json_dumps = pydantic_orjson_dumps
-
-
-# All dstack models inherit from pydantic-duality's DualBaseModel.
-# DualBaseModel creates two classes for the model:
-# one with extra = "forbid" (CoreModel/CoreModel.__request__),
-# and another with extra = "ignore" (CoreModel.__response__).
-# This allows to use the same model both for strict parsing of the user input and
-# for permissive parsing of the server responses.
-#
-# We define a func to generate CoreModel dynamically that can be used
-# to define custom Config for both __request__ and __response__ models.
-# Note: Defining config in the model class directly overrides
-# pydantic-duality's base config, breaking __response__.
-def generate_dual_core_model(
-    custom_config: Union[type, Mapping],
-) -> "type[CoreModel]":
-    class CoreModel(generate_dual_base_model(custom_config)):
-        def json(
-            self,
-            *,
-            include: Optional[IncludeExcludeType] = None,
-            exclude: Optional[IncludeExcludeType] = None,
-            by_alias: bool = False,
-            skip_defaults: Optional[bool] = None,  # ignore as it's deprecated
-            exclude_unset: bool = False,
-            exclude_defaults: bool = False,
-            exclude_none: bool = False,
-            encoder: Optional[Callable[[Any], Any]] = None,
-            models_as_dict: bool = True,  # does not seems to be needed by dstack or dependencies
-            **dumps_kwargs: Any,
-        ) -> str:
-            """
-            Override `json()` method so that it calls `dict()`.
-            Allows changing how models are serialized by overriding `dict()` only.
-            By default, `json()` won't call `dict()`, so changes applied in `dict()` won't take place.
-            """
-            data = self.dict(
-                by_alias=by_alias,
-                include=include,
-                exclude=exclude,
-                exclude_unset=exclude_unset,
-                exclude_defaults=exclude_defaults,
-                exclude_none=exclude_none,
-            )
-            if self.__custom_root_type__:
-                data = data["__root__"]
-            return self.__config__.json_dumps(data, default=encoder, **dumps_kwargs)
-
-    return CoreModel
-
-
-if TYPE_CHECKING:
-
-    class CoreModel(generate_dual_base_model(CoreConfig)):
-        pass
-else:
-    CoreModel = generate_dual_core_model(CoreConfig)
-
-
-class FrozenConfig(CoreConfig):
-    frozen = True
-
-
-FrozenCoreModel = generate_dual_core_model(FrozenConfig)
-
-
-class Duration(int):
+class CoreModel(BaseModel):
     """
-    Duration in seconds.
+    The base class for all dstack models.
+
+    Unknown fields are rejected, which is what makes `dstack apply` report a typo'd key in a
+    user's YAML and what makes the API reject an unexpected request body. Reading a stored blob
+    or a peer's response needs the opposite — see `validate_extra_ignore` below.
     """
 
-    @classmethod
-    def __get_validators__(cls):
-        yield cls.parse
+    model_config = ConfigDict(
+        extra="forbid",
+        # YAML numbers reach str fields as int/float, e.g. a `python: 3.10` style shorthand.
+        coerce_numbers_to_str=True,
+    )
 
-    @classmethod
-    def parse(cls, v: Union[int, str]) -> "Duration":
-        if isinstance(v, (int, float)):
-            return cls(v)
-        if isinstance(v, str):
-            try:
-                return cls(int(v))
-            except ValueError:
-                pass
-            regex = re.compile(r"(?P<amount>\d+) *(?P<unit>[smhdw])$")
-            re_match = regex.match(v)
-            if not re_match:
-                raise ValueError(f"Cannot parse the duration {v}")
-            amount, unit = int(re_match.group("amount")), re_match.group("unit")
-            multiplier = {
-                "s": 1,
-                "m": 60,
-                "h": 3600,
-                "d": 24 * 3600,
-                "w": 7 * 24 * 3600,
-            }[unit]
-            return cls(amount * multiplier)
-        raise ValueError(f"Cannot parse the duration {v}")
+
+class FrozenCoreModel(CoreModel):
+    model_config = ConfigDict(frozen=True)
+
+
+T = TypeVar("T")
+
+_type_adapters: dict[Any, TypeAdapter] = {}
+
+
+@overload
+def validate_extra_ignore(tp: type[T], obj: Any) -> T: ...
+
+
+@overload
+def validate_extra_ignore(tp: Any, obj: Any) -> Any: ...
+
+
+def validate_extra_ignore(tp: Any, obj: Any) -> Any:
+    """
+    Validate `obj` against `tp` with `extra="ignore"`, dropping unknown fields at every level.
+
+    This is the read path: anything decoded from a stored blob or from a peer's response goes
+    through here, so that a newer writer adding a field does not break an older reader.
+
+    `obj` may be an instance of a *different* model class, which is how the backend configurators
+    re-read an `AWSBackendConfigWithCreds` as an `AWSConfig`. v1's `parse_obj` accepted that
+    directly; v2 needs a dict, so dump first.
+    """
+    if isinstance(obj, BaseModel):
+        obj = obj.model_dump()
+    return _get_type_adapter(tp).validate_python(obj, extra="ignore")
+
+
+@overload
+def validate_json_extra_ignore(tp: type[T], data: Union[str, bytes]) -> T: ...
+
+
+@overload
+def validate_json_extra_ignore(tp: Any, data: Union[str, bytes]) -> Any: ...
+
+
+def validate_json_extra_ignore(tp: Any, data: Union[str, bytes]) -> Any:
+    """
+    The JSON-input counterpart of `validate_extra_ignore`, keeping native JSON parsing rather
+    than going through `json.loads` and then validating in Python mode.
+    """
+    return _get_type_adapter(tp).validate_json(data, extra="ignore")
+
+
+def _get_type_adapter(tp: Any) -> TypeAdapter:
+    # Constructing a TypeAdapter builds a schema and a validator, so reuse them per type.
+    try:
+        adapter = _type_adapters.get(tp)
+    except TypeError:
+        # An unhashable annotation cannot be cached. Rare enough not to matter.
+        return TypeAdapter(tp)
+    if adapter is None:
+        adapter = TypeAdapter(tp)
+        _type_adapters[tp] = adapter
+    return adapter
 
 
 class RegistryAuth(FrozenCoreModel):
diff --git a/src/dstack/_internal/core/models/configurations.py b/src/dstack/_internal/core/models/configurations.py
index 17cf55f71c..bb81c37ed0 100644
--- a/src/dstack/_internal/core/models/configurations.py
+++ b/src/dstack/_internal/core/models/configurations.py
@@ -5,29 +5,38 @@
 from pathlib import PurePosixPath
 from typing import Annotated, Any, Dict, List, Literal, Optional, Union
 
-import orjson
-from pydantic import Field, ValidationError, conint, constr, root_validator, validator
+from pydantic import (
+    BeforeValidator,
+    ConfigDict,
+    Field,
+    GetCoreSchemaHandler,
+    RootModel,
+    ValidationError,
+    ValidationInfo,
+    conint,
+    constr,
+    field_validator,
+    model_validator,
+)
+from pydantic_core import CoreSchema, core_schema
 from typing_extensions import Self
 
 from dstack._internal.core.errors import ConfigurationError
 from dstack._internal.core.models.common import (
-    CoreConfig,
+    JSON_SCHEMA_DIALECT,
     CoreModel,
-    Duration,
     EntityReference,
     RegistryAuth,
-    generate_dual_core_model,
+    validate_extra_ignore,
 )
+from dstack._internal.core.models.duration import Duration, parse_off_duration
 from dstack._internal.core.models.envs import Env
 from dstack._internal.core.models.files import FilePathMapping
 from dstack._internal.core.models.fleets import FleetConfiguration
 from dstack._internal.core.models.gateways import GatewayConfiguration
 from dstack._internal.core.models.profiles import (
     ProfileParams,
-    ProfileParamsConfig,
     SpotPolicy,
-    parse_duration,
-    parse_off_duration,
 )
 from dstack._internal.core.models.resources import Range, ResourcesSpec
 from dstack._internal.core.models.routers import AnyServiceRouterConfig, ReplicaGroupRouterConfig
@@ -44,10 +53,6 @@
 from dstack._internal.core.services import is_valid_replica_group_name
 from dstack._internal.proxy.gateway.const import SERVICE_SCALING_WINDOWS
 from dstack._internal.utils.common import has_duplicates, list_enum_values_for_annotation
-from dstack._internal.utils.json_schema import add_extra_schema_types
-from dstack._internal.utils.json_utils import (
-    pydantic_orjson_dumps_with_indent,
-)
 
 CommandsList = List[str]
 ValidPort = conint(gt=0, le=65536)
@@ -86,6 +91,7 @@ class PythonVersion(str, Enum):
     PY311 = "3.11"
     PY312 = "3.12"
     PY313 = "3.13"
+    PY314 = "3.14"
 
 
 class PortMapping(CoreModel):
@@ -167,6 +173,25 @@ class RepoSpec(CoreModel):
         ),
     ] = RepoExistsAction.ERROR
 
+    @classmethod
+    def __get_pydantic_core_schema__(
+        cls, source_type: Any, handler: GetCoreSchemaHandler
+    ) -> CoreSchema:
+        model_schema = handler(source_type)
+        return core_schema.no_info_before_validator_function(
+            cls._parse_shorthand,
+            model_schema,
+            json_schema_input_schema=core_schema.union_schema(
+                [model_schema, core_schema.str_schema()]
+            ),
+        )
+
+    @classmethod
+    def _parse_shorthand(cls, v: Any) -> Any:
+        if isinstance(v, str):
+            return cls.parse(v)
+        return v
+
     @classmethod
     def parse(cls, v: str) -> Self:
         is_url = False
@@ -193,15 +218,16 @@ def parse(cls, v: str) -> Self:
             return cls(local_path=parts[0], path=parts[1])
         raise ValueError(f"Invalid repo: {v}")
 
-    @root_validator
-    def validate_local_path_or_url(cls, values):
-        if values["local_path"] and values["url"]:
+    @model_validator(mode="after")
+    def validate_local_path_or_url(self) -> Self:
+        if self.local_path and self.url:
             raise ValueError("`local_path` and `url` are mutually exclusive")
-        if not values["local_path"] and not values["url"]:
+        if not self.local_path and not self.url:
             raise ValueError("Either `local_path` or `url` must be specified")
-        return values
+        return self
 
-    @validator("path")
+    @field_validator("path")
+    @classmethod
     def validate_path(cls, v: Optional[str]) -> Optional[str]:
         if v is None:
             return v
@@ -256,7 +282,8 @@ class ScalingSpec(CoreModel):
         ),
     ] = Duration.parse("10m")
 
-    @validator("window")
+    @field_validator("window")
+    @classmethod
     def validate_window(cls, v: Optional[Duration]) -> Optional[Duration]:
         if v is not None and v not in SERVICE_SCALING_WINDOWS:
             raise ValueError(f"Window must be one of: {ALLOWED_SCALING_WINDOWS_DESCRIPTION}")
@@ -273,7 +300,7 @@ class HeaderPartitioningKey(CoreModel):
         str,
         Field(
             description="Name of the header to use for partitioning",
-            regex=r"^[a-zA-Z0-9-_]+$",  # prevent Nginx config injection
+            pattern=r"^[a-zA-Z0-9-_]+$",  # prevent Nginx config injection
             max_length=500,  # chosen randomly, Nginx limit is higher
         ),
     ]
@@ -288,7 +315,7 @@ class RateLimit(CoreModel):
                 " If an incoming request matches several prefixes, the longest prefix is applied"
             ),
             max_length=4094,  # Nginx limit
-            regex=r"^/[^\s\\{}]*$",  # prevent Nginx config injection
+            pattern=r"^/[^\s\\{}]*$",  # prevent Nginx config injection
         ),
     ] = "/"
     key: Annotated[
@@ -349,20 +376,7 @@ class HTTPHeaderSpec(CoreModel):
     ]
 
 
-class ProbeConfigConfig(CoreConfig):
-    @staticmethod
-    def schema_extra(schema: Dict[str, Any]):
-        add_extra_schema_types(
-            schema["properties"]["timeout"],
-            extra_types=[{"type": "string"}],
-        )
-        add_extra_schema_types(
-            schema["properties"]["interval"],
-            extra_types=[{"type": "string"}],
-        )
-
-
-class ProbeConfig(generate_dual_core_model(ProbeConfigConfig)):
+class ProbeConfig(CoreModel):
     type: Annotated[
         Literal["http"],
         Field(description="The probe type. Must be `http`"),
@@ -381,7 +395,7 @@ class ProbeConfig(generate_dual_core_model(ProbeConfigConfig)):
     ] = None
     headers: Annotated[
         list[HTTPHeaderSpec],
-        Field(description="A list of HTTP headers to include in the request", max_items=16),
+        Field(description="A list of HTTP headers to include in the request", max_length=16),
     ] = []
     body: Annotated[
         Optional[str],
@@ -392,7 +406,7 @@ class ProbeConfig(generate_dual_core_model(ProbeConfigConfig)):
         ),
     ] = None
     timeout: Annotated[
-        Optional[int],
+        Optional[Duration],
         Field(
             description=(
                 f"Maximum amount of time the HTTP request is allowed to take. Defaults to `{DEFAULT_PROBE_TIMEOUT}s`"
@@ -400,7 +414,7 @@ class ProbeConfig(generate_dual_core_model(ProbeConfigConfig)):
         ),
     ] = None
     interval: Annotated[
-        Optional[int],
+        Optional[Duration],
         Field(
             description=(
                 "Minimum amount of time between the end of one probe execution"
@@ -430,25 +444,22 @@ class ProbeConfig(generate_dual_core_model(ProbeConfigConfig)):
         ),
     ] = None
 
-    @validator("timeout", pre=True)
-    def parse_timeout(cls, v: Optional[Union[int, str]]) -> Optional[int]:
-        if v is None:
-            return v
-        parsed = parse_duration(v)
-        if parsed < MIN_PROBE_TIMEOUT:
+    @field_validator("timeout")
+    @classmethod
+    def validate_timeout(cls, v: Optional[Duration]) -> Optional[Duration]:
+        if v is not None and v < MIN_PROBE_TIMEOUT:
             raise ValueError(f"Probe timeout cannot be shorter than {MIN_PROBE_TIMEOUT}s")
-        return parsed
+        return v
 
-    @validator("interval", pre=True)
-    def parse_interval(cls, v: Optional[Union[int, str]]) -> Optional[int]:
-        if v is None:
-            return v
-        parsed = parse_duration(v)
-        if parsed < MIN_PROBE_INTERVAL:
+    @field_validator("interval")
+    @classmethod
+    def validate_interval(cls, v: Optional[Duration]) -> Optional[Duration]:
+        if v is not None and v < MIN_PROBE_INTERVAL:
             raise ValueError(f"Probe interval cannot be shorter than {MIN_PROBE_INTERVAL}s")
-        return parsed
+        return v
 
-    @validator("url")
+    @field_validator("url")
+    @classmethod
     def validate_url(cls, v: Optional[str]) -> Optional[str]:
         if v is None:
             return v
@@ -460,25 +471,45 @@ def validate_url(cls, v: Optional[str]) -> Optional[str]:
             raise ValueError("Cannot contain non-printable characters")
         return v
 
-    @root_validator
-    def validate_body_matches_method(cls, values):
-        method: HTTPMethod = values["method"]
-        if values["body"] is not None and method in ["get", "head"]:
+    @model_validator(mode="after")
+    def validate_body_matches_method(self) -> Self:
+        method: HTTPMethod = self.method
+        if self.body is not None and method in ["get", "head"]:
             raise ValueError(f"Cannot set request body for the `{method}` method")
-        return values
+        return self
 
 
-class BaseRunConfigurationConfig(CoreConfig):
-    @staticmethod
-    def schema_extra(schema: Dict[str, Any]):
-        add_extra_schema_types(
-            schema["properties"]["volumes"]["items"],
-            extra_types=[{"type": "string"}],
-        )
-        add_extra_schema_types(
-            schema["properties"]["files"]["items"],
-            extra_types=[{"type": "string"}],
-        )
+def _parse_mount_point_shorthand(v: Union[MountPoint, str]) -> MountPoint:
+    if isinstance(v, str):
+        return parse_mount_point(v)
+    return v
+
+
+def _parse_port_shorthand(v: Union[int, str, PortMapping]) -> PortMapping:
+    if isinstance(v, int):
+        return PortMapping(local_port=v, container_port=v)
+    if isinstance(v, str):
+        return PortMapping.parse(v)
+    return v
+
+
+# `json_schema_input_type` keeps the shorthand visible in the generated JSON Schema, which used to
+# be patched in per field by a sibling config class.
+MountPointOrShorthand = Annotated[
+    MountPoint,
+    BeforeValidator(_parse_mount_point_shorthand, json_schema_input_type=Union[MountPoint, str]),
+]
+# Declared as `PortMapping` rather than the input union: that is what the value always is once
+# `_parse_port_shorthand` has run.
+PortMappingOrShorthand = Annotated[
+    PortMapping,
+    BeforeValidator(
+        _parse_port_shorthand,
+        json_schema_input_type=Union[
+            ValidPort, constr(pattern=r"^(?:[0-9]+|\*):[0-9]+$"), PortMapping
+        ],
+    ),
+]
 
 
 class BaseRunConfiguration(CoreModel):
@@ -575,7 +606,9 @@ class BaseRunConfiguration(CoreModel):
             ),
         ),
     ] = None
-    volumes: Annotated[List[MountPoint], Field(description="The volumes mount points")] = []
+    volumes: Annotated[
+        List[MountPointOrShorthand], Field(description="The volumes mount points")
+    ] = []
     docker: Annotated[
         Optional[bool],
         Field(
@@ -605,9 +638,10 @@ class BaseRunConfiguration(CoreModel):
     dev environments it runs right before `init`.
     """
 
-    @validator("python", pre=True, always=True)
-    def convert_python(cls, v, values) -> Optional[PythonVersion]:
-        if v is not None and values.get("image"):
+    @field_validator("python", mode="before")
+    @classmethod
+    def convert_python(cls, v, info: ValidationInfo) -> Optional[PythonVersion]:
+        if v is not None and info.data.get("image"):
             raise ValueError("`image` and `python` are mutually exclusive fields")
         if isinstance(v, float):
             v = str(v)
@@ -617,50 +651,36 @@ def convert_python(cls, v, values) -> Optional[PythonVersion]:
             return PythonVersion(v)
         return v
 
-    @validator("docker", pre=True, always=True)
-    def _docker(cls, v, values) -> Optional[bool]:
-        if v is True and values.get("image"):
+    @field_validator("docker", mode="before")
+    @classmethod
+    def _docker(cls, v, info: ValidationInfo) -> Optional[bool]:
+        if v is True and info.data.get("image"):
             raise ValueError("`image` and `docker` are mutually exclusive fields")
-        if v is True and values.get("python"):
+        if v is True and info.data.get("python"):
             raise ValueError("`python` and `docker` are mutually exclusive fields")
-        if v is True and values.get("nvcc"):
+        if v is True and info.data.get("nvcc"):
             raise ValueError("`nvcc` and `docker` are mutually exclusive fields")
         # Ideally, we'd like to also prohibit privileged=False when docker=True,
         #   but it's not possible to do so without breaking backwards compatibility.
         return v
 
-    @validator("volumes", each_item=True, pre=True)
-    def convert_volumes(cls, v: Union[MountPoint, str]) -> MountPoint:
-        if isinstance(v, str):
-            return parse_mount_point(v)
-        return v
-
-    @validator("files", each_item=True, pre=True)
-    def convert_files(cls, v: Union[FilePathMapping, str]) -> FilePathMapping:
-        if isinstance(v, str):
-            return FilePathMapping.parse(v)
-        return v
-
-    @validator("repos", pre=True, each_item=True)
-    def convert_repos(cls, v: Union[RepoSpec, str]) -> RepoSpec:
-        if isinstance(v, str):
-            return RepoSpec.parse(v)
-        return v
-
-    @validator("repos")
+    @field_validator("repos")
+    @classmethod
     def validate_repos(cls, v) -> RepoSpec:
         if len(v) > 1:
             raise ValueError("A maximum of one repo is currently supported")
         return v
 
-    @validator("user")
+    @field_validator("user")
+    @classmethod
     def validate_user(cls, v) -> Optional[str]:
         if v is None:
             return None
         UnixUser.parse(v)
         return v
 
-    @validator("shell")
+    @field_validator("shell")
+    @classmethod
     def validate_shell(cls, v) -> Optional[str]:
         if v is None:
             return None
@@ -674,32 +694,24 @@ def validate_shell(cls, v) -> Optional[str]:
 
 class ConfigurationWithPortsParams(CoreModel):
     ports: Annotated[
-        List[Union[ValidPort, constr(regex=r"^(?:[0-9]+|\*):[0-9]+$"), PortMapping]],
+        List[PortMappingOrShorthand],
         Field(description="Port numbers/mapping to expose"),
     ] = []
 
-    @validator("ports", each_item=True)
-    def convert_ports(cls, v) -> PortMapping:
-        if isinstance(v, int):
-            return PortMapping(local_port=v, container_port=v)
-        elif isinstance(v, str):
-            return PortMapping.parse(v)
-        return v
-
 
 class ConfigurationWithCommandsParams(CoreModel):
     commands: Annotated[CommandsList, Field(description="The shell commands to run")] = []
 
-    @root_validator
-    def check_image_or_commands_present(cls, values):
+    @model_validator(mode="after")
+    def check_image_or_commands_present(self) -> Self:
         # If replicas is list, skip validation - commands come from replica groups
-        replicas = values.get("replicas")
+        replicas = getattr(self, "replicas", None)
         if isinstance(replicas, list):
-            return values
+            return self
 
-        if not values.get("commands") and not values.get("image"):
+        if not self.commands and not getattr(self, "image", None):
             raise ValueError("Either `commands` or `image` must be set")
-        return values
+        return self
 
 
 class DevEnvironmentConfigurationParams(CoreModel):
@@ -717,7 +729,7 @@ class DevEnvironmentConfigurationParams(CoreModel):
     ] = None
     init: Annotated[CommandsList, Field(description="The shell commands to run on startup")] = []
     inactivity_duration: Annotated[
-        Optional[Union[Literal["off"], int, bool, str]],
+        Optional[int],
         Field(
             description=(
                 "The maximum amount of time the dev environment can be inactive"
@@ -732,7 +744,13 @@ class DevEnvironmentConfigurationParams(CoreModel):
         ),
     ] = None
 
-    @validator("inactivity_duration", pre=True, allow_reuse=True)
+    # Not `OptionalOffableDuration`: "off" collapses to `None` here rather than staying as the string.
+    @field_validator(
+        "inactivity_duration",
+        mode="before",
+        json_schema_input_type=Optional[Union[Literal["off"], int, bool, str]],
+    )
+    @classmethod
     def parse_inactivity_duration(
         cls, v: Optional[Union[Literal["off"], int, bool, str]]
     ) -> Optional[int]:
@@ -741,10 +759,10 @@ def parse_inactivity_duration(
             return v
         return None
 
-    @root_validator
-    def validate_ide_and_version(cls, values):
-        ide = values.get("ide")
-        version = values.get("version")
+    @model_validator(mode="after")
+    def validate_ide_and_version(self) -> Self:
+        ide = self.ide
+        version = self.version
         if version and ide is None:
             raise ValueError("`version` requires `ide` to be set")
         if ide == "windsurf" and version:
@@ -754,17 +772,7 @@ def validate_ide_and_version(cls, values):
                     f"Invalid Windsurf version format: `{version}`. "
                     "Expected format: `version@commit` (e.g., `1.106.0@8951cd3ad688e789573d7f51750d67ae4a0bea7d`)"
                 )
-        return values
-
-
-class DevEnvironmentConfigurationConfig(
-    ProfileParamsConfig,
-    BaseRunConfigurationConfig,
-):
-    @staticmethod
-    def schema_extra(schema: Dict[str, Any]):
-        ProfileParamsConfig.schema_extra(schema)
-        BaseRunConfigurationConfig.schema_extra(schema)
+        return self
 
 
 class DevEnvironmentConfiguration(
@@ -772,62 +780,38 @@ class DevEnvironmentConfiguration(
     BaseRunConfiguration,
     ConfigurationWithPortsParams,
     DevEnvironmentConfigurationParams,
-    generate_dual_core_model(DevEnvironmentConfigurationConfig),
 ):
     type: Literal["dev-environment"] = "dev-environment"
 
-    @validator("entrypoint")
+    @field_validator("entrypoint")
+    @classmethod
     def validate_entrypoint(cls, v: Optional[str]) -> Optional[str]:
         if v is not None:
             raise ValueError("entrypoint is not supported for dev-environment")
         return v
 
-    @root_validator
-    def validate_dstack_and_inactivity_duration(cls, values):
-        if values.get("dstack") and values.get("inactivity_duration") is not None:
+    @model_validator(mode="after")
+    def validate_dstack_and_inactivity_duration(self) -> Self:
+        if self.dstack and self.inactivity_duration is not None:
             # The persistent server connection counts as activity, so inactivity is never detected
             raise ValueError("`dstack` is not supported together with `inactivity_duration`")
-        return values
+        return self
 
 
 class TaskConfigurationParams(CoreModel):
     nodes: Annotated[int, Field(description="Number of nodes", ge=1)] = 1
 
 
-class TaskConfigurationConfig(
-    ProfileParamsConfig,
-    BaseRunConfigurationConfig,
-):
-    @staticmethod
-    def schema_extra(schema: Dict[str, Any]):
-        ProfileParamsConfig.schema_extra(schema)
-        BaseRunConfigurationConfig.schema_extra(schema)
-
-
 class TaskConfiguration(
     ProfileParams,
     BaseRunConfiguration,
     ConfigurationWithCommandsParams,
     ConfigurationWithPortsParams,
     TaskConfigurationParams,
-    generate_dual_core_model(TaskConfigurationConfig),
 ):
     type: Literal["task"] = "task"
 
 
-class ServiceConfigurationParamsConfig(CoreConfig):
-    @staticmethod
-    def schema_extra(schema: Dict[str, Any]):
-        add_extra_schema_types(
-            schema["properties"]["replicas"],
-            extra_types=[{"type": "integer"}, {"type": "string"}],
-        )
-        add_extra_schema_types(
-            schema["properties"]["model"],
-            extra_types=[{"type": "string"}],
-        )
-
-
 def _validate_replica_range(v: Range[int]) -> Range[int]:
     """Validate a Range[int] used for replica counts."""
     if v.max is None:
@@ -925,20 +909,23 @@ class ReplicaGroup(CoreModel):
         ),
     ] = None
 
-    @validator("name")
+    @field_validator("name")
+    @classmethod
     def validate_name(cls, v: Optional[str]) -> Optional[str]:
         if v is not None:
             if not is_valid_replica_group_name(v):
                 raise ValueError("Resource name should match regex '^[a-z0-9][a-z0-9-]{0,39}$'")
         return v
 
-    @validator("count")
+    @field_validator("count")
+    @classmethod
     def convert_count(cls, v: Range[int]) -> Range[int]:
         return _validate_replica_range(v)
 
-    @validator("python", pre=True, always=True)
-    def convert_python(cls, v, values) -> Optional[PythonVersion]:
-        if v is not None and values.get("image"):
+    @field_validator("python", mode="before")
+    @classmethod
+    def convert_python(cls, v, info: ValidationInfo) -> Optional[PythonVersion]:
+        if v is not None and info.data.get("image"):
             raise ValueError("`image` and `python` are mutually exclusive within a replica group")
         if isinstance(v, float):
             v = str(v)
@@ -948,45 +935,47 @@ def convert_python(cls, v, values) -> Optional[PythonVersion]:
             return PythonVersion(v)
         return v
 
-    @validator("docker", pre=True, always=True)
-    def _docker(cls, v, values) -> Optional[bool]:
-        if v is True and values.get("image"):
+    @field_validator("docker", mode="before")
+    @classmethod
+    def _docker(cls, v, info: ValidationInfo) -> Optional[bool]:
+        if v is True and info.data.get("image"):
             raise ValueError("`image` and `docker` are mutually exclusive within a replica group")
-        if v is True and values.get("python"):
+        if v is True and info.data.get("python"):
             raise ValueError("`python` and `docker` are mutually exclusive within a replica group")
-        if v is True and values.get("nvcc"):
+        if v is True and info.data.get("nvcc"):
             raise ValueError("`nvcc` and `docker` are mutually exclusive within a replica group")
         return v
 
-    @validator("privileged", pre=True, always=True)
-    def _privileged(cls, v, values) -> Optional[bool]:
+    @field_validator("privileged", mode="before")
+    @classmethod
+    def _privileged(cls, v, info: ValidationInfo) -> Optional[bool]:
         # Docker-in-docker requires privileged mode. The service level
         # cannot enforce this rule because its `privileged` field defaults
         # to `False` (existing backwards-compatibility constraint), so it
         # cannot distinguish "unset" from explicit `False`. At the group
         # level we keep `privileged` as `Optional[bool] = None`, so we can.
-        if v is False and values.get("docker") is True:
+        if v is False and info.data.get("docker") is True:
             raise ValueError(
                 "`privileged: false` is incompatible with `docker: true` within "
                 "a replica group (docker-in-docker requires privileged mode)"
             )
         return v
 
-    @root_validator()
-    def validate_scaling(cls, values):
-        scaling = values.get("scaling")
-        count = values.get("count")
+    @model_validator(mode="after")
+    def validate_scaling(self) -> Self:
+        scaling = self.scaling
+        count = self.count
         if count and count.min != count.max and not scaling:
             raise ValueError("When you set `count` to a range, ensure to specify `scaling`.")
         if count and count.min == count.max and scaling:
             raise ValueError("To use `scaling`, `count` must be set to a range.")
-        return values
+        return self
 
 
 class ServiceConfigurationParams(CoreModel):
     port: Annotated[
         # NOTE: it's a PortMapping for historical reasons. Only `port.container_port` is used.
-        Union[ValidPort, constr(regex=r"^[0-9]+:[0-9]+$"), PortMapping],
+        Union[ValidPort, constr(pattern=r"^[0-9]+:[0-9]+$"), PortMapping],
         Field(description="The port the application listens on"),
     ]
     gateway: Annotated[
@@ -998,6 +987,7 @@ class ServiceConfigurationParams(CoreModel):
             ]
         ],
         Field(
+            union_mode="left_to_right",  # preserving pydantic v1 parsing behavior
             description=(
                 "The name of the gateway. Specify boolean `false` to run without a gateway."
                 " Specify boolean `true` to run with the default gateway."
@@ -1077,7 +1067,8 @@ class ServiceConfigurationParams(CoreModel):
         ),
     ] = None
 
-    @validator("port")
+    @field_validator("port")
+    @classmethod
     def convert_port(cls, v) -> PortMapping:
         if isinstance(v, int):
             return PortMapping(local_port=80, container_port=v)
@@ -1085,13 +1076,15 @@ def convert_port(cls, v) -> PortMapping:
             return PortMapping.parse(v)
         return v
 
-    @validator("model", pre=True)
+    @field_validator("model", mode="before", json_schema_input_type=Optional[Union[AnyModel, str]])
+    @classmethod
     def convert_model(cls, v: Optional[Union[AnyModel, str]]) -> Optional[AnyModel]:
         if isinstance(v, str):
             return OpenAIChatModel(type="chat", name=v, format="openai")
         return v
 
-    @validator("rate_limits")
+    @field_validator("rate_limits")
+    @classmethod
     def validate_rate_limits(cls, v: list[RateLimit]) -> list[RateLimit]:
         counts = Counter(limit.prefix for limit in v)
         duplicates = [prefix for prefix, count in counts.items() if count > 1]
@@ -1102,7 +1095,8 @@ def validate_rate_limits(cls, v: list[RateLimit]) -> list[RateLimit]:
             )
         return v
 
-    @validator("probes")
+    @field_validator("probes")
+    @classmethod
     def validate_probes(cls, v: Optional[list[ProbeConfig]]) -> Optional[list[ProbeConfig]]:
         if v is None:
             return v
@@ -1114,7 +1108,8 @@ def validate_probes(cls, v: Optional[list[ProbeConfig]]) -> Optional[list[ProbeC
             raise ValueError("Probes must be unique")
         return v
 
-    @validator("gateway")
+    @field_validator("gateway")
+    @classmethod
     def validate_gateway(
         cls, v: Optional[Union[bool, EntityReference, str]]
     ) -> Optional[Union[bool, EntityReference]]:
@@ -1122,7 +1117,8 @@ def validate_gateway(
             return EntityReference.parse(v)
         return v
 
-    @validator("replicas")
+    @field_validator("replicas")
+    @classmethod
     def validate_replicas(
         cls, v: Optional[Union[Range[int], List[ReplicaGroup]]]
     ) -> Optional[Union[Range[int], List[ReplicaGroup]]]:
@@ -1150,10 +1146,10 @@ def validate_replicas(
                 )
         return v
 
-    @root_validator()
-    def validate_scaling(cls, values):
-        scaling = values.get("scaling")
-        replicas = values.get("replicas")
+    @model_validator(mode="after")
+    def validate_scaling(self) -> Self:
+        scaling = self.scaling
+        replicas = self.replicas
 
         if isinstance(replicas, Range):
             if replicas and replicas.min != replicas.max and not scaling:
@@ -1162,80 +1158,80 @@ def validate_scaling(cls, values):
                 )
             if replicas and replicas.min == replicas.max and scaling:
                 raise ValueError("To use `scaling`, `replicas` must be set to a range.")
-        return values
+        return self
 
-    @root_validator()
-    def validate_top_level_properties_with_replica_groups(cls, values):
+    @model_validator(mode="after")
+    def validate_top_level_properties_with_replica_groups(self) -> Self:
         """
         When replicas is a list of ReplicaGroup, forbid top-level scaling and commands.
         """
-        replicas = values.get("replicas")
+        replicas = self.replicas
 
         if not isinstance(replicas, list):
-            return values
+            return self
 
-        scaling = values.get("scaling")
+        scaling = self.scaling
         if scaling is not None:
             raise ValueError(
                 "Top-level `scaling` is not allowed when `replicas` is a list. "
                 "Specify `scaling` in each replica group instead."
             )
 
-        commands = values.get("commands", [])
+        commands = getattr(self, "commands", None)
         if commands:
             raise ValueError(
                 "Top-level `commands` is not allowed when `replicas` is a list. "
                 "Specify `commands` in each replica group instead."
             )
 
-        return values
+        return self
 
-    @root_validator()
-    def validate_no_mixed_service_and_group_container_fields(cls, values):
+    @model_validator(mode="after")
+    def validate_no_mixed_service_and_group_container_fields(self) -> Self:
         """
         When replicas is a list, certain fields may be set
         at the service level OR in replica groups, never both. Mixing is
         rejected — including partial mixing, where only some groups set a
         field the service also sets — because it leaves precedence ambiguous.
         """
-        replicas = values.get("replicas")
+        replicas = self.replicas
         if not isinstance(replicas, list):
-            return values
+            return self
 
         checks = [
             (
                 "image",
-                values.get("image") is not None,
+                getattr(self, "image", None) is not None,
                 lambda g: g.image is not None,
             ),
             (
                 "docker",
-                values.get("docker") is True,
+                getattr(self, "docker", None) is True,
                 lambda g: g.docker is not None,
             ),
             (
                 "privileged",
-                values.get("privileged") is True,
+                getattr(self, "privileged", None) is True,
                 lambda g: g.privileged is not None,
             ),
             (
                 "python",
-                values.get("python") is not None,
+                getattr(self, "python", None) is not None,
                 lambda g: g.python is not None,
             ),
             (
                 "nvcc",
-                values.get("nvcc") is True,
+                getattr(self, "nvcc", None) is True,
                 lambda g: g.nvcc is not None,
             ),
             (
                 "spot_policy",
-                values.get("spot_policy") is not None,
+                getattr(self, "spot_policy", None) is not None,
                 lambda g: g.spot_policy is not None,
             ),
             (
                 "reservation",
-                values.get("reservation") is not None,
+                getattr(self, "reservation", None) is not None,
                 lambda g: g.reservation is not None,
             ),
         ]
@@ -1250,29 +1246,74 @@ def validate_no_mixed_service_and_group_container_fields(cls, values):
                         f"place only — either at the service level (all groups "
                         f"inherit) or per group, but not both."
                     )
-        return values
+        return self
 
-    @root_validator()
-    def validate_no_conflicting_image_sources_across_levels(cls, values):
+    @model_validator(mode="after")
+    def validate_no_conflicting_image_sources_across_levels(self) -> Self:
         """
         Image-source fields (`image`, `docker`, `python`, `nvcc`) cannot
         be mixed across service and group levels in conflicting ways.
         """
-        replicas = values.get("replicas")
+        replicas = self.replicas
         if not isinstance(replicas, list):
-            return values
+            return self
 
         forbidden = [
-            ("image", values.get("image") is not None, "docker", lambda g: g.docker is not None),
-            ("image", values.get("image") is not None, "python", lambda g: g.python is not None),
-            ("image", values.get("image") is not None, "nvcc", lambda g: g.nvcc is not None),
-            ("docker", values.get("docker") is True, "image", lambda g: g.image is not None),
-            ("docker", values.get("docker") is True, "python", lambda g: g.python is not None),
-            ("docker", values.get("docker") is True, "nvcc", lambda g: g.nvcc is not None),
-            ("python", values.get("python") is not None, "image", lambda g: g.image is not None),
-            ("python", values.get("python") is not None, "docker", lambda g: g.docker is not None),
-            ("nvcc", values.get("nvcc") is True, "image", lambda g: g.image is not None),
-            ("nvcc", values.get("nvcc") is True, "docker", lambda g: g.docker is not None),
+            (
+                "image",
+                getattr(self, "image", None) is not None,
+                "docker",
+                lambda g: g.docker is not None,
+            ),
+            (
+                "image",
+                getattr(self, "image", None) is not None,
+                "python",
+                lambda g: g.python is not None,
+            ),
+            (
+                "image",
+                getattr(self, "image", None) is not None,
+                "nvcc",
+                lambda g: g.nvcc is not None,
+            ),
+            (
+                "docker",
+                getattr(self, "docker", None) is True,
+                "image",
+                lambda g: g.image is not None,
+            ),
+            (
+                "docker",
+                getattr(self, "docker", None) is True,
+                "python",
+                lambda g: g.python is not None,
+            ),
+            (
+                "docker",
+                getattr(self, "docker", None) is True,
+                "nvcc",
+                lambda g: g.nvcc is not None,
+            ),
+            (
+                "python",
+                getattr(self, "python", None) is not None,
+                "image",
+                lambda g: g.image is not None,
+            ),
+            (
+                "python",
+                getattr(self, "python", None) is not None,
+                "docker",
+                lambda g: g.docker is not None,
+            ),
+            ("nvcc", getattr(self, "nvcc", None) is True, "image", lambda g: g.image is not None),
+            (
+                "nvcc",
+                getattr(self, "nvcc", None) is True,
+                "docker",
+                lambda g: g.docker is not None,
+            ),
         ]
 
         for s_field, s_set, g_field, g_pred in forbidden:
@@ -1284,22 +1325,22 @@ def validate_no_conflicting_image_sources_across_levels(cls, values):
                         f"`{g_field}` in replica group(s) {conflicting}. "
                         f"These image-source fields are mutually exclusive."
                     )
-        return values
+        return self
 
-    @root_validator()
-    def validate_replica_groups_have_commands_or_image(cls, values):
+    @model_validator(mode="after")
+    def validate_replica_groups_have_commands_or_image(self) -> Self:
         """
         When replicas is a list, ensure each ReplicaGroup has something
         to run. Mirrors the service-level rule: either explicit
         `commands` or an `image` (group-level or service-level) is
         required.
         """
-        replicas = values.get("replicas")
+        replicas = self.replicas
 
         if not isinstance(replicas, list):
-            return values
+            return self
 
-        service_has_image = values.get("image") is not None
+        service_has_image = getattr(self, "image", None) is not None
 
         for group in replicas:
             if not group.commands and group.image is None and not service_has_image:
@@ -1309,13 +1350,13 @@ def validate_replica_groups_have_commands_or_image(cls, values):
                     "service level."
                 )
 
-        return values
+        return self
 
-    @root_validator()
-    def validate_at_most_one_router_replica_group(cls, values):
-        replicas = values.get("replicas")
+    @model_validator(mode="after")
+    def validate_at_most_one_router_replica_group(self) -> Self:
+        replicas = self.replicas
         if not isinstance(replicas, list):
-            return values
+            return self
         router_groups = [g for g in replicas if g.router is not None]
         if len(router_groups) > 1:
             raise ValueError("At most one replica group may specify `router`.")
@@ -1323,36 +1364,24 @@ def validate_at_most_one_router_replica_group(cls, values):
             router_group = router_groups[0]
             if router_group.count.min != 1 or router_group.count.max != 1:
                 raise ValueError("For now replica group with `router` must have `count: 1`.")
-        return values
+        return self
 
-    @root_validator()
-    def validate_replica_group_router_mutex(cls, values):
+    @model_validator(mode="after")
+    def validate_replica_group_router_mutex(self) -> Self:
         """
         When a replica group sets `router:`, service-level `router` must be omitted.
         (Gateway-level SGLang is rejected at service registration when a gateway is selected.)
         """
-        replicas = values.get("replicas")
+        replicas = self.replicas
         if not isinstance(replicas, list):
-            return values
+            return self
         if not any(g.router is not None for g in replicas):
-            return values
-        if values.get("router") is not None:
+            return self
+        if self.router is not None:
             raise ValueError(
                 "Service-Level router configuration is not allowed together with replica-group `router`."
             )
-        return values
-
-
-class ServiceConfigurationConfig(
-    ProfileParamsConfig,
-    BaseRunConfigurationConfig,
-    ServiceConfigurationParamsConfig,
-):
-    @staticmethod
-    def schema_extra(schema: Dict[str, Any]):
-        ProfileParamsConfig.schema_extra(schema)
-        BaseRunConfigurationConfig.schema_extra(schema)
-        ServiceConfigurationParamsConfig.schema_extra(schema)
+        return self
 
 
 class ServiceConfiguration(
@@ -1360,7 +1389,6 @@ class ServiceConfiguration(
     BaseRunConfiguration,
     ConfigurationWithCommandsParams,
     ServiceConfigurationParams,
-    generate_dual_core_model(ServiceConfigurationConfig),
 ):
     type: Literal["service"] = "service"
 
@@ -1396,16 +1424,13 @@ def replica_groups(self) -> List[ReplicaGroup]:
 AnyRunConfiguration = Union[DevEnvironmentConfiguration, TaskConfiguration, ServiceConfiguration]
 
 
-class RunConfiguration(CoreModel):
-    __root__: Annotated[
-        AnyRunConfiguration,
-        Field(discriminator="type"),
-    ]
+class RunConfiguration(RootModel[Annotated[AnyRunConfiguration, Field(discriminator="type")]]):
+    pass
 
 
 def parse_run_configuration(data: dict) -> AnyRunConfiguration:
     try:
-        conf = RunConfiguration.parse_obj(data).__root__
+        conf = RunConfiguration.model_validate(data).root
     except ValidationError as e:
         raise ConfigurationError(e)
     return conf
@@ -1428,7 +1453,20 @@ class ApplyConfigurationType(str, Enum):
 ]
 
 
-class BaseApplyConfiguration(CoreModel):
+_AnyBaseApplyConfiguration = Annotated[
+    Union[
+        # Final configurations
+        AnyRunConfiguration,
+        FleetConfiguration,
+        GatewayConfiguration,
+        # Base configurations (further parsing required to get a concrete AnyApplyConfiguration)
+        BaseVolumeConfiguration,
+    ],
+    Field(discriminator="type"),
+]
+
+
+class BaseApplyConfiguration(RootModel[_AnyBaseApplyConfiguration]):
     """
     `BaseApplyConfiguration` parses the configuration based on the `type` discriminator field,
     but further dispatching (reparsing) may be required if there is another discriminator field,
@@ -1438,28 +1476,16 @@ class BaseApplyConfiguration(CoreModel):
     Don't use this model directly, use `parse_apply_configuration()` instead.
     """
 
-    __root__: Annotated[
-        Union[
-            # Final configurations
-            AnyRunConfiguration,
-            FleetConfiguration,
-            GatewayConfiguration,
-            # Base configurations (further parsing required to get a concrete AnyApplyConfiguration)
-            BaseVolumeConfiguration,
-        ],
-        Field(discriminator="type"),
-    ]
-
 
 def parse_apply_configuration(data: dict) -> AnyApplyConfiguration:
     try:
         # First-pass parsing ignoring extra fields, to get the base (or final) configuration
-        conf = BaseApplyConfiguration.__response__.parse_obj(data).__root__
+        conf = validate_extra_ignore(BaseApplyConfiguration, data).root
         if not isinstance(conf, BaseVolumeConfiguration):
             # If it's a final configuration (currently, any configuration other than
             # BaseVolumeConfiguration), parse again rejecting extra fields
             # for validation purposes only and return the final configuration
-            _ = BaseApplyConfiguration.parse_obj(data).__root__
+            _ = BaseApplyConfiguration.model_validate(data).root
             return conf
     except ValidationError as e:
         raise ConfigurationError(e)
@@ -1475,19 +1501,14 @@ def parse_apply_configuration(data: dict) -> AnyApplyConfiguration:
 ]
 
 
-class DstackConfiguration(CoreModel):
-    __root__: Annotated[
-        AnyDstackConfiguration,
-        Field(discriminator="type"),
-    ]
+def _dstack_configuration_schema(schema: Dict[str, Any]) -> None:
+    schema["$schema"] = JSON_SCHEMA_DIALECT
+    # Allow additionalProperties so that vscode and others not supporting
+    # top-level oneOf do not warn about properties being invalid.
+    schema["additionalProperties"] = True
 
-    class Config(CoreConfig):
-        json_loads = orjson.loads
-        json_dumps = pydantic_orjson_dumps_with_indent
 
-        @staticmethod
-        def schema_extra(schema: Dict[str, Any]):
-            schema["$schema"] = "http://json-schema.org/draft-07/schema#"
-            # Allow additionalProperties so that vscode and others not supporting
-            # top-level oneOf do not warn about properties being invalid.
-            schema["additionalProperties"] = True
+class DstackConfiguration(
+    RootModel[Annotated[AnyDstackConfiguration, Field(discriminator="type")]]
+):
+    model_config = ConfigDict(json_schema_extra=_dstack_configuration_schema)
diff --git a/src/dstack/_internal/core/models/duration.py b/src/dstack/_internal/core/models/duration.py
new file mode 100644
index 0000000000..e063a60de2
--- /dev/null
+++ b/src/dstack/_internal/core/models/duration.py
@@ -0,0 +1,130 @@
+import re
+from typing import Any, Optional, Union
+
+from pydantic import BeforeValidator, GetCoreSchemaHandler, GetJsonSchemaHandler
+from pydantic.json_schema import JsonSchemaValue
+from pydantic_core import CoreSchema, core_schema
+from typing_extensions import Annotated, Literal, overload
+
+
+class Duration(int):
+    """
+    Duration in seconds.
+    """
+
+    @classmethod
+    def parse(cls, v: Union[int, str]) -> "Duration":
+        if isinstance(v, (int, float)):
+            return cls(v)
+        if isinstance(v, str):
+            try:
+                return cls(int(v))
+            except ValueError:
+                pass
+            regex = re.compile(r"(?P<amount>\d+) *(?P<unit>[smhdw])$")
+            re_match = regex.match(v)
+            if not re_match:
+                raise ValueError(f"Cannot parse the duration {v}")
+            amount, unit = int(re_match.group("amount")), re_match.group("unit")
+            multiplier = {
+                "s": 1,
+                "m": 60,
+                "h": 3600,
+                "d": 24 * 3600,
+                "w": 7 * 24 * 3600,
+            }[unit]
+            return cls(amount * multiplier)
+        raise ValueError(f"Cannot parse the duration {v}")
+
+    @classmethod
+    def __get_pydantic_core_schema__(
+        cls, source_type: Any, handler: GetCoreSchemaHandler
+    ) -> CoreSchema:
+        return core_schema.no_info_plain_validator_function(
+            cls.parse,
+            serialization=core_schema.plain_serializer_function_ser_schema(
+                int, return_schema=core_schema.int_schema()
+            ),
+        )
+
+    @classmethod
+    def __get_pydantic_json_schema__(
+        cls, schema: CoreSchema, handler: GetJsonSchemaHandler
+    ) -> JsonSchemaValue:
+        # A duration is accepted either as a number of seconds or as a shorthand string
+        # like `2h`, but it always serializes as a number of seconds.
+        if handler.mode == "validation":
+            return {"anyOf": [{"type": "integer"}, {"type": "string"}]}
+        return {"type": "integer"}
+
+
+@overload
+def parse_duration(v: None) -> None: ...
+
+
+@overload
+def parse_duration(v: Union[int, str]) -> int: ...
+
+
+def parse_duration(v: Optional[Union[int, str]]) -> Optional[int]:
+    if v is None:
+        return None
+    return Duration.parse(v)
+
+
+def parse_off_duration(v: Optional[Union[int, str, bool]]) -> Optional[Union[Literal["off"], int]]:
+    if v == "off" or v is False:
+        return "off"
+    if v is True or v is None:
+        return None
+    duration = parse_duration(v)
+    if duration < 0:
+        raise ValueError("Duration cannot be negative")
+    return duration
+
+
+def parse_idle_duration(v: Optional[Union[int, str, bool]]) -> Optional[int]:
+    # Differs from `parse_off_duration` to accept negative durations as `off`
+    # for backward compatibility.
+    if v == "off" or v is False or v == -1:
+        return -1
+    if v is True:
+        return None
+    return parse_duration(v)
+
+
+# Both include `None` in their own value domain rather than being wrapped in `Optional[...]` at the
+# field, which is why the names say so. `None` means "unspecified, use the default" and `true` is the
+# documented way to ask for it, so it belongs to the domain. It is also required mechanically: a
+# `BeforeValidator` nested inside an `Optional[...]` runs *after* the nullable check, so the `None`
+# it returns for `true` would then be rejected by the inner union. Use plain `Duration` for a
+# duration that is required.
+#
+# The parsed value is `int`, not `Duration`, even though the parse functions return `Duration`,
+# because code may assign non-`Duration` values directly. Switching to `Duration` would fail serialization in such cases.
+
+OptionalOffableDuration = Annotated[
+    Optional[Union[Literal["off"], int]],
+    BeforeValidator(parse_off_duration, json_schema_input_type=Optional[Union[int, str, bool]]),
+]
+"""
+A duration that can be switched off. Value domain: `None` (unspecified), `"off"`, or seconds.
+
+`false` and `"off"` both normalize to `"off"`; `true` and omission both normalize to `None`.
+Negative values are rejected — contrast `OptionalIdleDuration`.
+"""
+
+OptionalIdleDuration = Annotated[
+    Optional[int],
+    BeforeValidator(
+        parse_idle_duration,
+        json_schema_input_type=Optional[Union[Literal["off"], int, str, bool]],
+    ),
+]
+"""
+A duration whose "off" state is the sentinel `-1` rather than a string.
+
+`false`, `"off"` and `-1` all normalize to `-1`, and any negative value is accepted and left as
+is. That is deliberate and load-bearing: `-1` is what older clients and existing stored rows use
+to mean "off", so rejecting negatives here would break reads of data already in the database.
+"""
diff --git a/src/dstack/_internal/core/models/envs.py b/src/dstack/_internal/core/models/envs.py
index 5109c01d5d..92e1575a66 100644
--- a/src/dstack/_internal/core/models/envs.py
+++ b/src/dstack/_internal/core/models/envs.py
@@ -1,7 +1,7 @@
 import re
 from typing import Dict, Iterable, Iterator, List, Mapping, NamedTuple, Tuple, Union, cast
 
-from pydantic import BaseModel, Field, validator
+from pydantic import ConfigDict, Field, RootModel, field_validator
 from typing_extensions import Annotated, Self
 
 from dstack._internal.core.models.common import CoreModel
@@ -39,25 +39,34 @@ def parse(cls, v: str) -> Self:
         return cls(key, value)
 
 
-class Env(BaseModel):
+# Accepted either as a list of `VAR=value`/`VAR` strings or as a mapping. `validate_root` below
+# normalizes the list form into the mapping form, so the runtime value is always a dict.
+_EnvRoot = Union[
+    List[Annotated[str, Field(pattern=_ENV_STRING_REGEX)]],
+    Dict[str, Union[str, EnvSentinel]],
+]
+
+
+class Env(RootModel[_EnvRoot]):
     """
     Env represents a mapping of process environment variables, as in environ(7).
     Environment values may be omitted, in that case the :class:`EnvSentinel`
     object is used as a placeholder.
 
     To create an instance from a `dict[str, str]` or a `list[str]` use pydantic's
-    :meth:`BaseModel.parse_obj(dict | list)` method.
+    :meth:`BaseModel.model_validate(dict | list)` method.
 
-    NB: this is *NOT* a CoreModel, pydantic-duality, which is used as a base
-    for the CoreModel, doesn't play well with custom root models.
+    NB: this is *NOT* a CoreModel. `extra` is meaningless on a root model, but
+    `coerce_numbers_to_str` is not: without it `env: {PORT: 8080}` stops parsing, since
+    pydantic v2 does not coerce a YAML number to a str implicitly.
     """
 
-    __root__: Union[
-        List[Annotated[str, Field(regex=_ENV_STRING_REGEX)]],
-        Dict[str, Union[str, EnvSentinel]],
-    ] = {}
+    model_config = ConfigDict(coerce_numbers_to_str=True)
 
-    @validator("__root__")
+    root: _EnvRoot = {}
+
+    @field_validator("root")
+    @classmethod
     def validate_root(cls, v: Union[List[str], Dict[str, str]]) -> Dict[str, str]:
         if isinstance(v, list):
             d = {}
@@ -99,13 +108,13 @@ def __getitem__(self, item):
     def __setitem__(self, item, value):
         self._dict[item] = value
 
-    def copy(self, **kwargs) -> Self:
-        # Env.copy() is tricky because it copies only the hidden top-level {"__root__": {...}}
+    def model_copy(self, **kwargs) -> Self:
+        # Env.model_copy() is tricky because it copies only the hidden top-level {"root": {...}}
         # structure, not the actual nested dict representing the env itself.
-        # So we copy __root__ explicitly in case of a shallow copy.
-        new_copy = super().copy(**kwargs)
+        # So we copy root explicitly in case of a shallow copy.
+        new_copy = super().model_copy(**kwargs)
         if not kwargs.get("deep", False):
-            new_copy.__root__ = new_copy.__root__.copy()
+            new_copy.root = new_copy.root.copy()
         return new_copy
 
     def as_dict(self) -> Dict[str, str]:
@@ -146,4 +155,4 @@ def items(self) -> Iterable[Tuple[str, Union[str, EnvSentinel]]]:
     @property
     def _dict(self) -> Dict[str, Union[str, EnvSentinel]]:
         # this property is redundant for runtime and used for _proper_ type signature only
-        return cast(Dict, self.__root__)
+        return cast(Dict, self.root)
diff --git a/src/dstack/_internal/core/models/files.py b/src/dstack/_internal/core/models/files.py
index 2c82fd53b5..1db962165a 100644
--- a/src/dstack/_internal/core/models/files.py
+++ b/src/dstack/_internal/core/models/files.py
@@ -1,8 +1,10 @@
 import pathlib
 import string
+from typing import Any
 from uuid import UUID
 
-from pydantic import Field, validator
+from pydantic import Field, GetCoreSchemaHandler, field_validator
+from pydantic_core import CoreSchema, core_schema
 from typing_extensions import Annotated, Self
 
 from dstack._internal.core.models.common import CoreModel
@@ -33,6 +35,25 @@ class FilePathMapping(CoreModel):
         ),
     ]
 
+    @classmethod
+    def __get_pydantic_core_schema__(
+        cls, source_type: Any, handler: GetCoreSchemaHandler
+    ) -> CoreSchema:
+        model_schema = handler(source_type)
+        return core_schema.no_info_before_validator_function(
+            cls._parse_shorthand,
+            model_schema,
+            json_schema_input_schema=core_schema.union_schema(
+                [model_schema, core_schema.str_schema()]
+            ),
+        )
+
+    @classmethod
+    def _parse_shorthand(cls, v: Any) -> Any:
+        if isinstance(v, str):
+            return cls.parse(v)
+        return v
+
     @classmethod
     def parse(cls, v: str) -> Self:
         local_path: str
@@ -54,7 +75,8 @@ def parse(cls, v: str) -> Self:
             raise ValueError(f"invalid file path mapping: {v}")
         return cls(local_path=local_path, path=path)
 
-    @validator("path")
+    @field_validator("path")
+    @classmethod
     def validate_path(cls, v) -> str:
         # True for `C:/.*`, False otherwise, including `/abs/unix/path`, `rel\windows\path`, etc.
         if pathlib.PureWindowsPath(v).is_absolute():
diff --git a/src/dstack/_internal/core/models/fleets.py b/src/dstack/_internal/core/models/fleets.py
index ce636ba8de..963edbfd56 100644
--- a/src/dstack/_internal/core/models/fleets.py
+++ b/src/dstack/_internal/core/models/fleets.py
@@ -4,17 +4,24 @@
 from enum import Enum
 from typing import Any, Dict, List, Optional, Union
 
-from pydantic import Field, root_validator, validator
-from typing_extensions import Annotated, Literal
+from pydantic import (
+    ConfigDict,
+    Field,
+    SerializerFunctionWrapHandler,
+    field_validator,
+    model_serializer,
+    model_validator,
+)
+from typing_extensions import Annotated, Literal, Self
 
 from dstack._internal.core.backends.profile_options import AnyBackendProfileOptions
 from dstack._internal.core.models.backends.base import BackendType
 from dstack._internal.core.models.common import (
     ApplyAction,
-    CoreConfig,
     CoreModel,
-    generate_dual_core_model,
+    drop_merged_profile,
 )
+from dstack._internal.core.models.duration import OptionalIdleDuration
 from dstack._internal.core.models.envs import Env
 from dstack._internal.core.models.instances import Instance, InstanceOfferWithAvailability, SSHKey
 from dstack._internal.core.models.profiles import (
@@ -22,12 +29,10 @@
     ProfileParams,
     ProfileRetry,
     SpotPolicy,
-    parse_idle_duration,
     validate_backend_options,
 )
 from dstack._internal.core.models.resources import ResourcesSpec
 from dstack._internal.utils.common import list_enum_values_for_annotation
-from dstack._internal.utils.json_schema import add_extra_schema_types
 from dstack._internal.utils.tags import tags_validator
 
 
@@ -80,7 +85,7 @@ class SSHHostParams(CoreModel):
     ssh_key: Optional[SSHKey] = None
 
     blocks: Annotated[
-        Optional[Union[Literal["auto"], int]],
+        Optional[Union[Literal["auto"], Annotated[int, Field(ge=1)]]],
         Field(
             description=(
                 "The amount of blocks to split the instance into, a number or `auto`."
@@ -88,11 +93,11 @@ class SSHHostParams(CoreModel):
                 " The number of GPUs and CPUs must be divisible by the number of blocks."
                 " Defaults to the top-level `blocks` value"
             ),
-            ge=1,
         ),
     ] = None
 
-    @validator("internal_ip")
+    @field_validator("internal_ip")
+    @classmethod
     def validate_internal_ip(cls, value):
         if value is None:
             return value
@@ -134,7 +139,8 @@ class SSHParams(CoreModel):
         ),
     ] = None
 
-    @validator("network")
+    @field_validator("network")
+    @classmethod
     def validate_network(cls, value):
         if value is None:
             return value
@@ -169,16 +175,17 @@ class FleetNodesSpec(CoreModel):
         ),
     ] = None
 
-    def dict(self, *args, **kwargs) -> Dict:
-        # super() does not work with pydantic-duality
-        res = CoreModel.dict(self, *args, **kwargs)
+    @model_serializer(mode="wrap")
+    def _serialize(self, handler: SerializerFunctionWrapHandler) -> Dict[str, Any]:
+        res = handler(self)
         # For backward compatibility with old clients
         # that do not ignore extra fields due to https://github.com/dstackai/dstack/issues/3066
         if "target" in res and res["target"] == res["min"]:
             del res["target"]
         return res
 
-    @root_validator(pre=True)
+    @model_validator(mode="before")
+    @classmethod
     def set_min_and_target_defaults(cls, values):
         min_ = values.get("min")
         target = values.get("target")
@@ -188,24 +195,25 @@ def set_min_and_target_defaults(cls, values):
             values["target"] = values["min"]
         return values
 
-    @validator("min")
+    @field_validator("min")
+    @classmethod
     def validate_min(cls, v: int) -> int:
         if v < 0:
             raise ValueError("min cannot be negative")
         return v
 
-    @root_validator(skip_on_failure=True)
-    def _post_validate_ranges(cls, values):
-        min_ = values["min"]
-        target = values["target"]
-        max_ = values.get("max")
+    @model_validator(mode="after")
+    def _post_validate_ranges(self) -> Self:
+        min_ = self.min
+        target = self.target
+        max_ = self.max
         if target < min_:
             raise ValueError("target must not be be less than min")
         if max_ is not None and max_ < min_:
             raise ValueError("max must not be less than min")
         if max_ is not None and max_ < target:
             raise ValueError("max must not be less than target")
-        return values
+        return self
 
 
 class CommonFleetConfigurationProps(CoreModel):
@@ -216,7 +224,7 @@ class CommonFleetConfigurationProps(CoreModel):
         Field(description="The placement of instances: `any` or `cluster`"),
     ] = None
     blocks: Annotated[
-        Union[Literal["auto"], int],
+        Union[Literal["auto"], Annotated[int, Field(ge=1)]],
         Field(
             description=(
                 "The amount of blocks to split the instance into, a number or `auto`."
@@ -224,7 +232,6 @@ class CommonFleetConfigurationProps(CoreModel):
                 " The number of GPUs and CPUs must be divisible by the number of blocks."
                 " Defaults to `1`, i.e. do not split"
             ),
-            ge=1,
         ),
     ] = 1
 
@@ -285,7 +292,7 @@ class BackendFleetConfiguraionProps(CoreModel):
         Field(description="The maximum instance price per hour, in dollars", gt=0.0),
     ] = None
     idle_duration: Annotated[
-        Optional[int],
+        OptionalIdleDuration,
         Field(
             description=(
                 "Time to wait before terminating idle instances."
@@ -310,7 +317,10 @@ class BackendFleetConfiguraionProps(CoreModel):
         Field(description="Backend-specific options, applied only to offers from that backend"),
     ] = None
 
-    @validator("nodes", pre=True)
+    @field_validator(
+        "nodes", mode="before", json_schema_input_type=Optional[Union[FleetNodesSpec, int, str]]
+    )
+    @classmethod
     def parse_nodes(cls, v: Optional[Union[dict, str]]) -> Optional[dict]:
         if isinstance(v, str) and ".." in v:
             v = v.replace(" ", "")
@@ -320,26 +330,8 @@ def parse_nodes(cls, v: Optional[Union[dict, str]]) -> Optional[dict]:
             return dict(min=v, max=v)
         return v
 
-    _validate_idle_duration = validator("idle_duration", pre=True, allow_reuse=True)(
-        parse_idle_duration
-    )
-    _validate_tags = validator("tags", pre=True, allow_reuse=True)(tags_validator)
-    _validate_backend_options = validator("backend_options", allow_reuse=True)(
-        validate_backend_options
-    )
-
-
-class BackendFleetConfigurationPropsConfig(CoreConfig):
-    @staticmethod
-    def schema_extra(schema: Dict[str, Any]):
-        add_extra_schema_types(
-            schema["properties"]["nodes"],
-            extra_types=[{"type": "integer"}, {"type": "string"}],
-        )
-        add_extra_schema_types(
-            schema["properties"]["idle_duration"],
-            extra_types=[{"type": "string"}],
-        )
+    _validate_tags = field_validator("tags", mode="before")(tags_validator)
+    _validate_backend_options = field_validator("backend_options")(validate_backend_options)
 
 
 class SSHFleetConfigurationProps(CoreModel):
@@ -353,17 +345,10 @@ class SSHFleetConfigurationProps(CoreModel):
     ] = Env()
 
 
-class FleetConfigurationConfig(BackendFleetConfigurationPropsConfig):
-    @staticmethod
-    def schema_extra(schema: dict[str, Any]):
-        BackendFleetConfigurationPropsConfig.schema_extra(schema)
-
-
 class FleetConfiguration(
     SSHFleetConfigurationProps,
     BackendFleetConfiguraionProps,
     CommonFleetConfigurationProps,
-    generate_dual_core_model(FleetConfigurationConfig),
 ):
     pass
 
@@ -371,7 +356,6 @@ class FleetConfiguration(
 class BackendFleetConfiguration(
     BackendFleetConfiguraionProps,
     CommonFleetConfigurationProps,
-    generate_dual_core_model(BackendFleetConfigurationPropsConfig),
 ):
     """For the documentation only"""
 
@@ -383,14 +367,9 @@ class SSHFleetConfiguration(
     """For the documentation only"""
 
 
-class FleetSpecConfig(CoreConfig):
-    @staticmethod
-    def schema_extra(schema: Dict[str, Any]):
-        prop = schema.get("properties", {})
-        prop.pop("merged_profile", None)
-
+class FleetSpec(CoreModel):
+    model_config = ConfigDict(json_schema_extra=drop_merged_profile)
 
-class FleetSpec(generate_dual_core_model(FleetSpecConfig)):
     configuration: FleetConfiguration
     configuration_path: Optional[str] = None
     profile: Profile
@@ -399,20 +378,23 @@ class FleetSpec(generate_dual_core_model(FleetSpecConfig)):
     autocreated: bool = False
     """Deprecated. Kept for deserialization of old client requests and existing DB records.
     """
-    # TODO: make `merged_profile` a computed field after migrating to Pydantic v2.
+    # TODO: consider a `property` or `cached_property` instead of an excluded field.
     merged_profile: Annotated[Profile, Field(exclude=True)] = None
     """`merged_profile` stores profile parameters merged from `profile` and `configuration`.
     Read profile parameters from `merged_profile` instead of `profile` directly.
     """
 
-    @root_validator
+    @model_validator(mode="before")
+    @classmethod
     def _merged_profile(cls, values) -> Dict:
         try:
-            merged_profile = Profile.parse_obj(values["profile"])
-            conf = FleetConfiguration.parse_obj(values["configuration"])
+            # Copy first: `model_validate` returns the *same* instance for a same-class input, so
+            # the `setattr` loop below would otherwise mutate the caller's profile in place.
+            merged_profile = Profile.model_validate(values["profile"]).model_copy(deep=True)
+            conf = FleetConfiguration.model_validate(values["configuration"])
         except KeyError:
             raise ValueError("Missing profile or configuration")
-        for key in ProfileParams.__fields__:
+        for key in ProfileParams.model_fields:
             conf_val = getattr(conf, key, None)
             if conf_val is not None:
                 setattr(merged_profile, key, conf_val)
diff --git a/src/dstack/_internal/core/models/gateways.py b/src/dstack/_internal/core/models/gateways.py
index b4d2598c18..89da495743 100644
--- a/src/dstack/_internal/core/models/gateways.py
+++ b/src/dstack/_internal/core/models/gateways.py
@@ -3,7 +3,7 @@
 from enum import Enum
 from typing import Dict, Optional, Union
 
-from pydantic import Field, validator
+from pydantic import Field, RootModel, field_validator
 from typing_extensions import Annotated, Literal
 
 from dstack._internal.core.models.backends.base import BackendType
@@ -49,11 +49,8 @@ class ACMGatewayCertificate(CoreModel):
 AnyGatewayCertificate = Union[LetsEncryptGatewayCertificate, ACMGatewayCertificate]
 
 
-class GatewayCertificate(CoreModel):
-    __root__: Annotated[
-        AnyGatewayCertificate,
-        Field(discriminator="type"),
-    ]
+class GatewayCertificate(RootModel[Annotated[AnyGatewayCertificate, Field(discriminator="type")]]):
+    pass
 
 
 class GatewayConfiguration(CoreModel):
@@ -120,7 +117,7 @@ class GatewayConfiguration(CoreModel):
         ),
     ] = None
 
-    _validate_tags = validator("tags", pre=True, allow_reuse=True)(tags_validator)
+    _validate_tags = field_validator("tags", mode="before")(tags_validator)
 
 
 class GatewaySpec(CoreModel):
diff --git a/src/dstack/_internal/core/models/instances.py b/src/dstack/_internal/core/models/instances.py
index cb7656f15c..38fc803108 100644
--- a/src/dstack/_internal/core/models/instances.py
+++ b/src/dstack/_internal/core/models/instances.py
@@ -4,7 +4,7 @@
 from uuid import UUID
 
 import gpuhunt
-from pydantic import Field, root_validator
+from pydantic import Field, model_validator
 
 from dstack._internal.core.models.backends.base import BackendType
 from dstack._internal.core.models.common import (
@@ -28,7 +28,8 @@ class Gpu(CoreModel):
     `assert gpu.vendor is not None` should be a safe type narrowing.
     """
 
-    @root_validator(pre=True)
+    @model_validator(mode="before")
+    @classmethod
     def validate_name_and_vendor(cls, values):
         is_tpu = False
         name = values.get("name")
@@ -204,7 +205,7 @@ def with_availability(self, **kwargs) -> "InstanceOfferWithAvailability":
         """Convert to InstanceOfferWithAvailability without re-serializing/re-validating fields.
         The result shares nested objects with self. This is generally safe because callers
         discard the original InstanceOffer after conversion."""
-        return InstanceOfferWithAvailability.construct(**self.__dict__, **kwargs)
+        return InstanceOfferWithAvailability.model_construct(**self.__dict__, **kwargs)
 
 
 class InstanceOfferWithAvailability(InstanceOffer):
diff --git a/src/dstack/_internal/core/models/profiles.py b/src/dstack/_internal/core/models/profiles.py
index 7a448486df..5e64ea2f0e 100644
--- a/src/dstack/_internal/core/models/profiles.py
+++ b/src/dstack/_internal/core/models/profiles.py
@@ -1,26 +1,33 @@
 from enum import Enum
-from typing import Any, Dict, List, Optional, Union, overload
-
-import orjson
-from pydantic import Field, root_validator, validator
-from typing_extensions import Annotated, Literal
+from typing import Any, Dict, List, Optional, Union
+
+from pydantic import (
+    AfterValidator,
+    BeforeValidator,
+    ConfigDict,
+    Field,
+    field_validator,
+    model_validator,
+)
+from typing_extensions import Annotated, Self
 
 from dstack._internal.core.backends.profile_options import AnyBackendProfileOptions
 from dstack._internal.core.models.backends.base import BackendType
 from dstack._internal.core.models.common import (
-    CoreConfig,
+    JSON_SCHEMA_DIALECT,
     CoreModel,
-    Duration,
     EntityReference,
-    generate_dual_core_model,
+)
+from dstack._internal.core.models.duration import (
+    Duration,
+    OptionalIdleDuration,
+    OptionalOffableDuration,
 )
 from dstack._internal.utils.common import list_enum_values_for_annotation
 from dstack._internal.utils.cron import validate_cron
-from dstack._internal.utils.json_schema import add_extra_schema_types
-from dstack._internal.utils.json_utils import pydantic_orjson_dumps_with_indent
 from dstack._internal.utils.tags import tags_validator
 
-DEFAULT_RETRY_DURATION = 3600
+DEFAULT_RETRY_DURATION = Duration(3600)
 
 DEFAULT_RUN_TERMINATION_IDLE_TIME = 5 * 60  # 5 minutes
 DEFAULT_FLEET_TERMINATION_IDLE_TIME = 72 * 60 * 60  # 3 days
@@ -55,51 +62,6 @@ class StopCriteria(str, Enum):
     MASTER_DONE = "master-done"
 
 
-@overload
-def parse_duration(v: None) -> None: ...
-
-
-@overload
-def parse_duration(v: Union[int, str]) -> int: ...
-
-
-def parse_duration(v: Optional[Union[int, str]]) -> Optional[int]:
-    if v is None:
-        return None
-    return Duration.parse(v)
-
-
-def parse_max_duration(v: Optional[Union[int, str, bool]]) -> Optional[Union[Literal["off"], int]]:
-    return parse_off_duration(v)
-
-
-def parse_stop_duration(
-    v: Optional[Union[int, str, bool]],
-) -> Optional[Union[Literal["off"], int]]:
-    return parse_off_duration(v)
-
-
-def parse_off_duration(v: Optional[Union[int, str, bool]]) -> Optional[Union[Literal["off"], int]]:
-    if v == "off" or v is False:
-        return "off"
-    if v is True or v is None:
-        return None
-    duration = parse_duration(v)
-    if duration < 0:
-        raise ValueError("Duration cannot be negative")
-    return duration
-
-
-def parse_idle_duration(v: Optional[Union[int, str, bool]]) -> Optional[int]:
-    # Differs from `parse_off_duration` to accept negative durations as `off`
-    # for backward compatibility.
-    if v == "off" or v is False or v == -1:
-        return -1
-    if v is True:
-        return None
-    return parse_duration(v)
-
-
 def validate_backend_options(
     v: Optional[List["AnyBackendProfileOptions"]],
 ) -> Optional[List["AnyBackendProfileOptions"]]:
@@ -119,16 +81,7 @@ class RetryEvent(str, Enum):
     ERROR = "error"
 
 
-class ProfileRetryConfig(CoreConfig):
-    @staticmethod
-    def schema_extra(schema: Dict[str, Any]):
-        add_extra_schema_types(
-            schema["properties"]["duration"],
-            extra_types=[{"type": "string"}],
-        )
-
-
-class ProfileRetry(generate_dual_core_model(ProfileRetryConfig)):
+class ProfileRetry(CoreModel):
     on_events: Annotated[
         Optional[List[RetryEvent]],
         Field(
@@ -140,7 +93,7 @@ class ProfileRetry(generate_dual_core_model(ProfileRetryConfig)):
         ),
     ] = None
     duration: Annotated[
-        Optional[int],
+        Optional[Duration],
         Field(
             description=(
                 "The maximum period of retrying the run, e.g., `4h` or `1d`."
@@ -150,28 +103,18 @@ class ProfileRetry(generate_dual_core_model(ProfileRetryConfig)):
         ),
     ] = None
 
-    _validate_duration = validator("duration", pre=True, allow_reuse=True)(parse_duration)
-
-    @root_validator
-    def _validate_fields(cls, values):
-        on_events = values.get("on_events", None)
-        if on_events is not None and len(values["on_events"]) == 0:
+    @model_validator(mode="after")
+    def _validate_fields(self) -> Self:
+        on_events = self.on_events
+        if on_events is not None and len(self.on_events) == 0:
             raise ValueError("`on_events` cannot be empty")
-        return values
+        return self
 
 
-class UtilizationPolicyConfig(CoreConfig):
-    @staticmethod
-    def schema_extra(schema: Dict[str, Any]):
-        add_extra_schema_types(
-            schema["properties"]["time_window"],
-            extra_types=[{"type": "string"}],
-        )
+MIN_UTILIZATION_TIME_WINDOW = "5m"
 
 
-class UtilizationPolicy(generate_dual_core_model(UtilizationPolicyConfig)):
-    _min_time_window = "5m"
-
+class UtilizationPolicy(CoreModel):
     min_gpu_utilization: Annotated[
         int,
         Field(
@@ -185,20 +128,20 @@ class UtilizationPolicy(generate_dual_core_model(UtilizationPolicyConfig)):
         ),
     ]
     time_window: Annotated[
-        int,
+        Duration,
         Field(
             description=(
                 "The time window of metric samples taking into account to measure utilization"
-                f" (e.g., `30m`, `1h`). Minimum is `{_min_time_window}`"
+                f" (e.g., `30m`, `1h`). Minimum is `{MIN_UTILIZATION_TIME_WINDOW}`"
             )
         ),
     ]
 
-    @validator("time_window", pre=True)
-    def validate_time_window(cls, v: Union[int, str]) -> int:
-        v = parse_duration(v)
-        if v < parse_duration(cls._min_time_window):
-            raise ValueError(f"Minimum time_window is {cls._min_time_window}")
+    @field_validator("time_window")
+    @classmethod
+    def validate_time_window(cls, v: Duration) -> Duration:
+        if v < Duration.parse(MIN_UTILIZATION_TIME_WINDOW):
+            raise ValueError(f"Minimum time_window is {MIN_UTILIZATION_TIME_WINDOW}")
         return v
 
 
@@ -212,7 +155,8 @@ class Schedule(CoreModel):
         ),
     ]
 
-    @validator("cron")
+    @field_validator("cron")
+    @classmethod
     def _validate_cron(cls, v: Union[List[str], str]) -> List[str]:
         if isinstance(v, str):
             values = [v]
@@ -250,16 +194,7 @@ def _parse_fleet_instance_selector_fleet(v: Any) -> Any:
     return v
 
 
-class FleetInstanceSelectorConfig(CoreConfig):
-    @staticmethod
-    def schema_extra(schema: Dict[str, Any]):
-        add_extra_schema_types(
-            schema["properties"]["fleet"],
-            extra_types=[{"type": "string", "minLength": 1}],
-        )
-
-
-class FleetInstanceSelector(generate_dual_core_model(FleetInstanceSelectorConfig)):
+class FleetInstanceSelector(CoreModel):
     fleet: Annotated[
         EntityReference,
         Field(
@@ -272,9 +207,9 @@ class FleetInstanceSelector(generate_dual_core_model(FleetInstanceSelectorConfig
     ]
     instance: Annotated[int, Field(description="The fleet instance number", ge=0)]
 
-    _validate_fleet = validator("fleet", pre=True, allow_reuse=True)(
-        _parse_fleet_instance_selector_fleet
-    )
+    _validate_fleet = field_validator(
+        "fleet", mode="before", json_schema_input_type=Union[EntityReference, str]
+    )(_parse_fleet_instance_selector_fleet)
 
 
 InstanceSelector = Union[InstanceNameSelector, InstanceHostnameSelector, FleetInstanceSelector]
@@ -286,25 +221,17 @@ def parse_instance_selector(v: Union[InstanceSelector, str]) -> InstanceSelector
     return v
 
 
-class ProfileParamsConfig(CoreConfig):
-    @staticmethod
-    def schema_extra(schema: Dict[str, Any]):
-        add_extra_schema_types(
-            schema["properties"]["max_duration"],
-            extra_types=[{"type": "boolean"}, {"type": "string"}],
-        )
-        add_extra_schema_types(
-            schema["properties"]["stop_duration"],
-            extra_types=[{"type": "boolean"}, {"type": "string"}],
-        )
-        add_extra_schema_types(
-            schema["properties"]["idle_duration"],
-            extra_types=[{"type": "string"}],
-        )
-        add_extra_schema_types(
-            schema["properties"]["instances"]["items"],
-            extra_types=[{"type": "string", "minLength": 1}],
-        )
+FleetReferenceOrShorthand = Annotated[
+    Union[
+        EntityReference,
+        str,  # For server response compatibility with pre-0.20.14 clients
+    ],
+    AfterValidator(EntityReference.parse),
+]
+InstanceSelectorOrShorthand = Annotated[
+    InstanceSelector,
+    BeforeValidator(parse_instance_selector, json_schema_input_type=Union[InstanceSelector, str]),
+]
 
 
 class ProfileParams(CoreModel):
@@ -354,7 +281,7 @@ class ProfileParams(CoreModel):
         Field(description="The policy for resubmitting the run. Defaults to `false`"),
     ] = None
     max_duration: Annotated[
-        Optional[Union[Literal["off"], int]],
+        OptionalOffableDuration,
         Field(
             description=(
                 "The maximum duration of a run (e.g., `2h`, `1d`, etc)"
@@ -365,7 +292,7 @@ class ProfileParams(CoreModel):
         ),
     ] = None
     stop_duration: Annotated[
-        Optional[Union[Literal["off"], int]],
+        OptionalOffableDuration,
         Field(
             description=(
                 "The maximum duration of a run graceful stopping."
@@ -390,7 +317,7 @@ class ProfileParams(CoreModel):
         ),
     ] = None
     idle_duration: Annotated[
-        Optional[int],
+        OptionalIdleDuration,
         Field(
             description=(
                 "Time to wait before terminating idle instances."
@@ -431,14 +358,7 @@ class ProfileParams(CoreModel):
         Field(description=("The schedule for starting the run at specified time")),
     ] = None
     fleets: Annotated[
-        Optional[
-            list[
-                Union[
-                    EntityReference,
-                    str,  # For server response compatibility with pre-0.20.14 clients
-                ]
-            ]
-        ],
+        Optional[list[FleetReferenceOrShorthand]],
         Field(
             description=(
                 "The fleets considered for reuse."
@@ -448,7 +368,7 @@ class ProfileParams(CoreModel):
         ),
     ] = None
     instances: Annotated[
-        Optional[List[InstanceSelector]],
+        Optional[List[InstanceSelectorOrShorthand]],
         Field(
             description=(
                 "The specific fleet instances to consider for reuse."
@@ -456,7 +376,7 @@ class ProfileParams(CoreModel):
                 " `name`, `hostname`, or `fleet` and `instance`."
                 " When set, the run is only placed on matching existing instances."
             ),
-            min_items=1,
+            min_length=1,
         ),
     ] = None
     tags: Annotated[
@@ -474,23 +394,8 @@ class ProfileParams(CoreModel):
         Field(description="Backend-specific options, applied only to offers from that backend"),
     ] = None
 
-    _validate_max_duration = validator("max_duration", pre=True, allow_reuse=True)(
-        parse_max_duration
-    )
-    _validate_stop_duration = validator("stop_duration", pre=True, allow_reuse=True)(
-        parse_stop_duration
-    )
-    _validate_idle_duration = validator("idle_duration", pre=True, allow_reuse=True)(
-        parse_idle_duration
-    )
-    _validate_fleets = validator("fleets", allow_reuse=True, each_item=True)(EntityReference.parse)
-    _validate_instances = validator("instances", pre=True, allow_reuse=True, each_item=True)(
-        parse_instance_selector
-    )
-    _validate_tags = validator("tags", pre=True, allow_reuse=True)(tags_validator)
-    _validate_backend_options = validator("backend_options", allow_reuse=True)(
-        validate_backend_options
-    )
+    _validate_tags = field_validator("tags", mode="before")(tags_validator)
+    _validate_backend_options = field_validator("backend_options")(validate_backend_options)
 
 
 class ProfileProps(CoreModel):
@@ -505,27 +410,16 @@ class ProfileProps(CoreModel):
     ] = False
 
 
-class ProfileConfig(ProfileParamsConfig):
-    @staticmethod
-    def schema_extra(schema: Dict[str, Any]):
-        ProfileParamsConfig.schema_extra(schema)
-
-
 class Profile(
     ProfileProps,
     ProfileParams,
-    generate_dual_core_model(ProfileConfig),
 ):
     pass
 
 
-class ProfilesConfigConfig(CoreConfig):
-    json_loads = orjson.loads
-    json_dumps = pydantic_orjson_dumps_with_indent
-    schema_extra = {"$schema": "http://json-schema.org/draft-07/schema#"}
-
+class ProfilesConfig(CoreModel):
+    model_config = ConfigDict(json_schema_extra={"$schema": JSON_SCHEMA_DIALECT})
 
-class ProfilesConfig(generate_dual_core_model(ProfilesConfigConfig)):
     profiles: List[Profile]
 
     def default(self) -> Optional[Profile]:
diff --git a/src/dstack/_internal/core/models/repos/remote.py b/src/dstack/_internal/core/models/repos/remote.py
index f613e18221..917969e7c5 100644
--- a/src/dstack/_internal/core/models/repos/remote.py
+++ b/src/dstack/_internal/core/models/repos/remote.py
@@ -3,11 +3,11 @@
 import subprocess
 import time
 from dataclasses import dataclass
-from typing import Annotated, Any, BinaryIO, Callable, Dict, Optional, Union, cast
+from typing import Annotated, BinaryIO, Callable, Dict, Optional, Union, cast
 
 import git
 import pydantic
-from pydantic import Field
+from pydantic import Field, TypeAdapter
 from typing_extensions import Literal
 
 from dstack._internal.core.deprecated import Deprecated
@@ -17,7 +17,7 @@
     RepoGitError,
     RepoInvalidGitRepositoryError,
 )
-from dstack._internal.core.models.common import CoreConfig, generate_dual_core_model
+from dstack._internal.core.models.common import CoreModel
 from dstack._internal.core.models.repos.base import BaseRepoInfo, Repo
 from dstack._internal.utils.hash import get_sha256, slugify
 from dstack._internal.utils.logging import get_logger
@@ -29,27 +29,14 @@
 SCP_LOCATION_REGEX = re.compile(r"(?P<user>[^/]+)@(?P<host>[^/]+?):(?P<path>.+)", re.IGNORECASE)
 
 
-class RemoteRepoCredsConfig(CoreConfig):
-    @staticmethod
-    def schema_extra(schema: Dict[str, Any]):
-        pass
-
-
-class RemoteRepoCreds(generate_dual_core_model(RemoteRepoCredsConfig)):
+class RemoteRepoCreds(CoreModel):
     clone_url: str
     private_key: Optional[str] = None
     oauth_token: Optional[str] = None
 
 
-class RemoteRepoInfoConfig(CoreConfig):
-    @staticmethod
-    def schema_extra(schema: Dict[str, Any]):
-        pass
-
-
 class RemoteRepoInfo(
     BaseRepoInfo,
-    generate_dual_core_model(RemoteRepoInfoConfig),
 ):
     repo_type: Literal["remote"] = "remote"
     repo_name: str
@@ -266,6 +253,22 @@ def get(self) -> bytes:
         return self.buffer.getvalue()
 
 
+HTTPS_DEFAULT_PORT = 443
+
+
+def _explicit_port(port: Optional[int], default: int) -> Optional[int]:
+    """
+    The port only if it was written explicitly.
+
+    pydantic v2's URL types fill in the scheme's default port, where v1 left `port` as `None`
+    unless the URL spelled it out. Without this, every https repo URL would be rebuilt as
+    `https://github.com:443/...`.
+    """
+    if port is None or port == default:
+        return None
+    return port
+
+
 @dataclass
 class GitRepoURL:
     """
@@ -287,7 +290,7 @@ def parse(
         get_ssh_config: Callable[[str], Dict[str, str]] = lambda host: {},
     ) -> "GitRepoURL":
         try:
-            url = pydantic.parse_obj_as(pydantic.AnyUrl, value)
+            url = TypeAdapter(pydantic.AnyUrl).validate_python(value)
         except pydantic.ValidationError:
             url = scp_location_to_ssh_url(value)
 
@@ -300,7 +303,7 @@ def parse(
             return GitRepoURL(
                 ssh_user=ssh_config.get("user"),
                 host=url.host.lower(),
-                https_port=url.port,
+                https_port=_explicit_port(url.port, default=HTTPS_DEFAULT_PORT),
                 ssh_port=ssh_config.get("port"),
                 path=url.path or "/",
                 original_host=url.host.lower(),
@@ -308,7 +311,7 @@ def parse(
 
         if url.scheme.lower() == "ssh":
             return GitRepoURL(
-                ssh_user=url.user or ssh_config.get("user"),
+                ssh_user=url.username or ssh_config.get("user"),
                 host=ssh_config.get("hostname", "").lower() or url.host.lower(),
                 https_port=None,
                 ssh_port=url.port or ssh_config.get("port"),
@@ -349,7 +352,7 @@ def scp_location_to_ssh_url(scp_location: str) -> Optional[pydantic.AnyHttpUrl]:
         return None
     user, host, path = match.group("user"), match.group("host"), match.group("path")
     try:
-        return pydantic.parse_obj_as(pydantic.AnyUrl, f"ssh://{user}@{host}/{path}")
+        return TypeAdapter(pydantic.AnyUrl).validate_python(f"ssh://{user}@{host}/{path}")
     except pydantic.ValidationError:
         return None
 
diff --git a/src/dstack/_internal/core/models/resources.py b/src/dstack/_internal/core/models/resources.py
index 72febf8eb9..92f7ac4c63 100644
--- a/src/dstack/_internal/core/models/resources.py
+++ b/src/dstack/_internal/core/models/resources.py
@@ -3,13 +3,24 @@
 from typing import Any, Dict, Generic, List, Optional, Tuple, TypeVar, Union
 
 import gpuhunt
-from pydantic import Field, parse_obj_as, root_validator, validator
-from pydantic.generics import GenericModel
+from pydantic import (
+    BaseModel,
+    ConfigDict,
+    Field,
+    GetCoreSchemaHandler,
+    GetJsonSchemaHandler,
+    SerializerFunctionWrapHandler,
+    Tag,
+    field_validator,
+    model_serializer,
+    model_validator,
+)
+from pydantic.json_schema import JsonSchemaValue
+from pydantic_core import CoreSchema, core_schema
 from typing_extensions import Annotated
 
-from dstack._internal.core.models.common import CoreConfig, CoreModel, generate_dual_core_model
+from dstack._internal.core.models.common import CoreModel
 from dstack._internal.utils.common import pretty_resources
-from dstack._internal.utils.json_schema import add_extra_schema_types
 from dstack._internal.utils.logging import get_logger
 
 logger = get_logger(__name__)
@@ -17,18 +28,29 @@
 
 T = TypeVar("T", bound=Union[int, float])
 
+# The shorthand forms these types accept in addition to their declared shape. Declared on the
+# type so that every field using it reports the same thing in the generated JSON Schema, instead
+# of each field restating it in a sibling config class.
+_INT_OR_STR_INPUT = [core_schema.int_schema(), core_schema.str_schema()]
+
+
+class Range(BaseModel, Generic[T]):
+    model_config = ConfigDict(extra="forbid")
 
-class Range(GenericModel, Generic[T]):
     min: Optional[T] = None
     max: Optional[T] = None
 
-    class Config:
-        extra = "forbid"
-
     @classmethod
-    def __get_validators__(cls):
-        yield cls._parse
-        yield cls.validate
+    def __get_pydantic_core_schema__(
+        cls, source_type: Any, handler: GetCoreSchemaHandler
+    ) -> CoreSchema:
+        model_schema = handler(source_type)
+        return core_schema.no_info_before_validator_function(
+            cls._parse,
+            model_schema,
+            # A range is also written as `8`, `2..8`, or `16GB..`.
+            json_schema_input_schema=core_schema.union_schema([model_schema, *_INT_OR_STR_INPUT]),
+        )
 
     @classmethod
     def _parse(cls, v: Any) -> Any:
@@ -40,16 +62,13 @@ def _parse(cls, v: Any) -> Any:
             return dict(min=v, max=v)
         return v
 
-    @root_validator()
-    def _post_validate(cls, values):
-        min = values.get("min")
-        max = values.get("max")
-
-        if min is None and max is None:
+    @model_validator(mode="after")
+    def _post_validate(self) -> "Range[T]":
+        if self.min is None and self.max is None:
             raise ValueError("Invalid empty range: ..")
-        if min is not None and max is not None and min > max:
-            raise ValueError(f"Invalid range order: {min}..{max}")
-        return values
+        if self.min is not None and self.max is not None and self.min > self.max:
+            raise ValueError(f"Invalid range order: {self.min}..{self.max}")
+        return self
 
     def __str__(self) -> str:
         min = self.min if self.min is not None else ""
@@ -80,10 +99,6 @@ class Memory(float):
     Memory size in gigabytes as a float number. Supported units: MB, GB, TB.
     """
 
-    @classmethod
-    def __get_validators__(cls):
-        yield cls.parse
-
     @classmethod
     def parse(cls, v: Any) -> "Memory":
         if isinstance(v, (float, int)):
@@ -102,12 +117,29 @@ def parse(cls, v: Any) -> "Memory":
     def __repr__(self):
         return f"{self:g}GB"
 
+    @classmethod
+    def __get_pydantic_core_schema__(
+        cls, source_type: Any, handler: GetCoreSchemaHandler
+    ) -> CoreSchema:
+        return core_schema.no_info_plain_validator_function(
+            cls.parse,
+            serialization=core_schema.plain_serializer_function_ser_schema(
+                float, return_schema=core_schema.float_schema()
+            ),
+        )
 
-class ComputeCapability(Tuple[int, int]):
     @classmethod
-    def __get_validators__(cls):
-        yield cls.validate
+    def __get_pydantic_json_schema__(
+        cls, schema: CoreSchema, handler: GetJsonSchemaHandler
+    ) -> JsonSchemaValue:
+        # A memory size is accepted as a number of gigabytes or as `16GB`/`512MB`/`1TB`,
+        # and always serializes as a number of gigabytes.
+        if handler.mode == "validation":
+            return {"anyOf": [{"type": "number"}, {"type": "integer"}, {"type": "string"}]}
+        return {"type": "number"}
+
 
+class ComputeCapability(Tuple[int, int]):
     @classmethod
     def validate(cls, v: Any) -> Tuple[int, int]:
         if isinstance(v, float):
@@ -123,22 +155,34 @@ def validate(cls, v: Any) -> Tuple[int, int]:
     def __str__(self):
         return f"{self[0]}.{self[1]}"
 
+    @classmethod
+    def __get_pydantic_core_schema__(
+        cls, source_type: Any, handler: GetCoreSchemaHandler
+    ) -> CoreSchema:
+        return core_schema.no_info_plain_validator_function(
+            cls.validate,
+            serialization=core_schema.plain_serializer_function_ser_schema(
+                list, return_schema=core_schema.list_schema(core_schema.int_schema())
+            ),
+        )
+
+    @classmethod
+    def __get_pydantic_json_schema__(
+        cls, schema: CoreSchema, handler: GetJsonSchemaHandler
+    ) -> JsonSchemaValue:
+        serialized = {"type": "array", "items": {"type": "integer"}}
+        if handler.mode == "validation":
+            # Written as `7.5` (a YAML float), as `"7.5"`, or as a two-element sequence.
+            return {"anyOf": [{"type": "number"}, {"type": "string"}, serialized]}
+        return serialized
+
 
 DEFAULT_CPU_COUNT = Range[int](min=2)
 DEFAULT_MEMORY_SIZE = Range[Memory](min=Memory.parse("8GB"))
 DEFAULT_GPU_COUNT = Range[int](min=1)
 
 
-class CPUSpecConfig(CoreConfig):
-    @staticmethod
-    def schema_extra(schema: Dict[str, Any]):
-        add_extra_schema_types(
-            schema["properties"]["count"],
-            extra_types=[{"type": "integer"}, {"type": "string"}],
-        )
-
-
-class CPUSpec(generate_dual_core_model(CPUSpecConfig)):
+class CPUSpec(CoreModel):
     arch: Annotated[
         Optional[gpuhunt.CPUArchitecture],
         Field(description="The CPU architecture, one of: `x86`, `arm`"),
@@ -146,9 +190,16 @@ class CPUSpec(generate_dual_core_model(CPUSpecConfig)):
     count: Annotated[Range[int], Field(description="The number of CPU cores")] = DEFAULT_CPU_COUNT
 
     @classmethod
-    def __get_validators__(cls):
-        yield cls.parse
-        yield cls.validate
+    def __get_pydantic_core_schema__(
+        cls, source_type: Any, handler: GetCoreSchemaHandler
+    ) -> CoreSchema:
+        model_schema = handler(source_type)
+        return core_schema.no_info_before_validator_function(
+            cls.parse,
+            model_schema,
+            # Also written as a count (`8`) or a `<arch>:<count>` shorthand (`arm:8`).
+            json_schema_input_schema=core_schema.union_schema([model_schema, *_INT_OR_STR_INPUT]),
+        )
 
     @classmethod
     def parse(cls, v: Any) -> Any:
@@ -176,11 +227,17 @@ def parse(cls, v: Any) -> Any:
         # Range and min/max dict - for backward compatibility
         if isinstance(v, Range):
             return {"arch": None, "count": v}
-        if isinstance(v, Mapping) and v.keys() == {"min", "max"}:
+        # A subset rather than exactly {"min", "max"}: `ResourcesSpec` serializes `cpu` down to its
+        # count for old clients, and under `exclude_none=True` that leaves just `{"min": ...}`.
+        # Requiring both keys made the round trip land on the `Range[int]` arm of `ResourcesSpec.cpu`
+        # instead of coming back as a `CPUSpec`. `arch`/`count` are the only `CPUSpec` fields, so a
+        # mapping of min/max is unambiguously a range.
+        if isinstance(v, Mapping) and v and v.keys() <= {"min", "max"}:
             return {"arch": None, "count": v}
         return v
 
-    @validator("arch", pre=True)
+    @field_validator("arch", mode="before")
+    @classmethod
     def _validate_arch(cls, v: Any) -> Any:
         if v is None:
             return None
@@ -191,28 +248,7 @@ def _validate_arch(cls, v: Any) -> Any:
         return v
 
 
-class GPUSpecConfig(CoreConfig):
-    @staticmethod
-    def schema_extra(schema: Dict[str, Any]):
-        add_extra_schema_types(
-            schema["properties"]["count"],
-            extra_types=[{"type": "integer"}, {"type": "string"}],
-        )
-        add_extra_schema_types(
-            schema["properties"]["name"],
-            extra_types=[{"type": "string"}],
-        )
-        add_extra_schema_types(
-            schema["properties"]["memory"],
-            extra_types=[{"type": "integer"}, {"type": "string"}],
-        )
-        add_extra_schema_types(
-            schema["properties"]["total_memory"],
-            extra_types=[{"type": "integer"}, {"type": "string"}],
-        )
-
-
-class GPUSpec(generate_dual_core_model(GPUSpecConfig)):
+class GPUSpec(CoreModel):
     vendor: Annotated[
         Optional[gpuhunt.AcceleratorVendor],
         Field(
@@ -241,9 +277,16 @@ class GPUSpec(generate_dual_core_model(GPUSpecConfig)):
     ] = None
 
     @classmethod
-    def __get_validators__(cls):
-        yield cls.parse
-        yield cls.validate
+    def __get_pydantic_core_schema__(
+        cls, source_type: Any, handler: GetCoreSchemaHandler
+    ) -> CoreSchema:
+        model_schema = handler(source_type)
+        return core_schema.no_info_before_validator_function(
+            cls.parse,
+            model_schema,
+            # Also written as a count (`8`) or a `:`-separated shorthand (`A100:8`, `nvidia:16GB`).
+            json_schema_input_schema=core_schema.union_schema([model_schema, *_INT_OR_STR_INPUT]),
+        )
 
     @classmethod
     def parse(cls, v: Any) -> Any:
@@ -280,7 +323,8 @@ def parse(cls, v: Any) -> Any:
             return spec
         return v
 
-    @validator("name", pre=True)
+    @field_validator("name", mode="before", json_schema_input_type=Optional[Union[List[str], str]])
+    @classmethod
     def _validate_name(cls, v: Any) -> Any:
         if v is None:
             return None
@@ -297,7 +341,8 @@ def _validate_name(cls, v: Any) -> Any:
             logger.warning("`tpu-` prefix is deprecated, specify gpu_vendor instead")
         return validated
 
-    @validator("vendor", pre=True)
+    @field_validator("vendor", mode="before")
+    @classmethod
     def _validate_vendor(
         cls, v: Union[str, gpuhunt.AcceleratorVendor, None]
     ) -> Optional[gpuhunt.AcceleratorVendor]:
@@ -307,7 +352,9 @@ def _validate_vendor(
             return v
         if isinstance(v, str):
             return cls._vendor_from_string(v)
-        raise TypeError(f"Unsupported type: {v!r}")
+        # A TypeError raised inside a validator is no longer converted to a ValidationError
+        # in pydantic v2, so it would escape as a 500 instead of a 422.
+        raise ValueError(f"Unsupported type: {v!r}")
 
     @classmethod
     def _vendor_from_string(cls, v: str) -> gpuhunt.AcceleratorVendor:
@@ -322,22 +369,20 @@ def _vendor_from_string(cls, v: str) -> gpuhunt.AcceleratorVendor:
 DEFAULT_GPU_SPEC = GPUSpec(count=Range[int](min=0, max=None))
 
 
-class DiskSpecConfig(CoreConfig):
-    @staticmethod
-    def schema_extra(schema: Dict[str, Any]):
-        add_extra_schema_types(
-            schema["properties"]["size"],
-            extra_types=[{"type": "integer"}, {"type": "string"}],
-        )
-
-
-class DiskSpec(generate_dual_core_model(DiskSpecConfig)):
+class DiskSpec(CoreModel):
     size: Annotated[Range[Memory], Field(description="Disk size")]
 
     @classmethod
-    def __get_validators__(cls):
-        yield cls._parse
-        yield cls.validate
+    def __get_pydantic_core_schema__(
+        cls, source_type: Any, handler: GetCoreSchemaHandler
+    ) -> CoreSchema:
+        model_schema = handler(source_type)
+        return core_schema.no_info_before_validator_function(
+            cls._parse,
+            model_schema,
+            # Also written as a bare size (`100GB`).
+            json_schema_input_schema=core_schema.union_schema([model_schema, *_INT_OR_STR_INPUT]),
+        )
 
     @classmethod
     def _parse(cls, v: Any) -> Any:
@@ -349,36 +394,21 @@ def _parse(cls, v: Any) -> Any:
 DEFAULT_DISK = DiskSpec(size=Range[Memory](min=Memory.parse("100GB"), max=None))
 
 
-class ResourcesSpecConfig(CoreConfig):
-    @staticmethod
-    def schema_extra(schema: Dict[str, Any]):
-        add_extra_schema_types(
-            schema["properties"]["cpu"],
-            extra_types=[{"type": "integer"}, {"type": "string"}],
-        )
-        add_extra_schema_types(
-            schema["properties"]["memory"],
-            extra_types=[{"type": "integer"}, {"type": "string"}],
-        )
-        add_extra_schema_types(
-            schema["properties"]["shm_size"],
-            extra_types=[{"type": "integer"}, {"type": "string"}],
-        )
-        add_extra_schema_types(
-            schema["properties"]["gpu"],
-            extra_types=[{"type": "integer"}, {"type": "string"}],
-        )
-        add_extra_schema_types(
-            schema["properties"]["disk"],
-            extra_types=[{"type": "integer"}, {"type": "string"}],
-        )
-
-
-class ResourcesSpec(generate_dual_core_model(ResourcesSpecConfig)):
+class ResourcesSpec(CoreModel):
     # TODO: remove `Range[int]` in 0.20. It is kept only for backward compatibility.
-    cpu: Annotated[Union[CPUSpec, Range[int]], Field(description="The CPU requirements")] = (
-        CPUSpec()
-    )
+    cpu: Annotated[
+        Union[
+            # `Tag` only names the arm in validation errors. Without it the `loc` of a bad `cpu`
+            # spells out the whole wrapped schema —
+            # `cpu.function-before[parse(), function-before[parse(), ... CPUSpec]].count` — which
+            # is what `dstack apply` shows the user.
+            Annotated[CPUSpec, Tag("CPUSpec")],
+            Annotated[Range[int], Tag("Range[int]")],
+        ],
+        # `CPUSpec` and `Range[int]` both accept a bare int/str, so the arm has to be picked by
+        # declaration order rather than by pydantic v2's "smart" union resolution.
+        Field(description="The CPU requirements", union_mode="left_to_right"),
+    ] = CPUSpec()
     memory: Annotated[Range[Memory], Field(description="The RAM size (e.g., `8GB`)")] = (
         DEFAULT_MEMORY_SIZE
     )
@@ -406,7 +436,7 @@ def unconstrained(cls) -> "ResourcesSpec":
 
     def pretty_format(self) -> str:
         # TODO: Remove in 0.20. Use self.cpu directly
-        cpu = parse_obj_as(CPUSpec, self.cpu)
+        cpu = CPUSpec.model_validate(self.cpu)
         resources: Dict[str, Any] = dict(cpu_arch=cpu.arch, cpus=cpu.count, memory=self.memory)
         if self.gpu:
             gpu = self.gpu
@@ -423,15 +453,15 @@ def pretty_format(self) -> str:
         res = pretty_resources(**resources)
         return res
 
-    def dict(self, *args, **kwargs) -> Dict:
-        # super() does not work with pydantic-duality
-        res = CoreModel.dict(self, *args, **kwargs)
+    @model_serializer(mode="wrap")
+    def _serialize(self, handler: SerializerFunctionWrapHandler) -> Dict[str, Any]:
+        res = handler(self)
         self._update_serialized_cpu(res)
         return res
 
     # TODO: Remove in 0.20. Added for backward compatibility.
     def _update_serialized_cpu(self, values: Dict):
-        cpu = values["cpu"]
+        cpu = values.get("cpu")
         if cpu:
             arch = cpu.get("arch")
             count = cpu.get("count")
diff --git a/src/dstack/_internal/core/models/runs.py b/src/dstack/_internal/core/models/runs.py
index 611ba87118..a292a928b8 100644
--- a/src/dstack/_internal/core/models/runs.py
+++ b/src/dstack/_internal/core/models/runs.py
@@ -3,18 +3,17 @@
 from typing import Any, Dict, List, Literal, Optional
 from urllib.parse import urlparse
 
-from pydantic import UUID4, Field, root_validator
-from typing_extensions import Annotated
+from pydantic import UUID4, ConfigDict, Field, model_validator
+from typing_extensions import Annotated, Self
 
 from dstack._internal.core.backends.profile_options import AnyBackendProfileOptions
 from dstack._internal.core.models.backends.base import BackendType
 from dstack._internal.core.models.common import (
     ApplyAction,
-    CoreConfig,
     CoreModel,
     NetworkMode,
     RegistryAuth,
-    generate_dual_core_model,
+    drop_merged_profile,
 )
 from dstack._internal.core.models.configurations import (
     DEFAULT_PROBE_METHOD,
@@ -519,14 +518,9 @@ class Job(CoreModel):
     job_connection_info: Optional[JobConnectionInfo] = None
 
 
-class RunSpecConfig(CoreConfig):
-    @staticmethod
-    def schema_extra(schema: Dict[str, Any]):
-        prop = schema.get("properties", {})
-        prop.pop("merged_profile", None)
+class RunSpec(CoreModel):
+    model_config = ConfigDict(json_schema_extra=drop_merged_profile)
 
-
-class RunSpec(generate_dual_core_model(RunSpecConfig)):
     # TODO: consider removing `run_name` here because it is already passed in `configuration`.
     run_name: Annotated[
         Optional[str],
@@ -588,23 +582,26 @@ class RunSpec(generate_dual_core_model(RunSpecConfig)):
             " Can be empty only before the run is submitted."
         ),
     ] = None
-    # TODO: make `merged_profile` a computed field after migrating to Pydantic v2.
+    # TODO: consider a `property` or `cached_property` instead of an excluded field.
     merged_profile: Annotated[Profile, Field(exclude=True)] = None
     """`merged_profile` stores profile parameters merged from `profile` and `configuration`.
     Read profile parameters from `merged_profile` instead of `profile` directly.
     """
 
-    @root_validator
+    @model_validator(mode="before")
+    @classmethod
     def _merged_profile(cls, values) -> Dict:
         if values.get("profile") is None:
             merged_profile = Profile(name="default")
         else:
-            merged_profile = Profile.parse_obj(values["profile"])
+            # Copy first: `model_validate` returns the *same* instance for a same-class input, so
+            # the `setattr` loop below would otherwise mutate the caller's profile in place.
+            merged_profile = Profile.model_validate(values["profile"]).model_copy(deep=True)
         try:
-            conf = RunConfiguration.parse_obj(values["configuration"]).__root__
+            conf = RunConfiguration.model_validate(values["configuration"]).root
         except KeyError:
             raise ValueError("Missing configuration")
-        for key in ProfileParams.__fields__:
+        for key in ProfileParams.model_fields:
             conf_val = getattr(conf, key, None)
             if conf_val is not None:
                 setattr(merged_profile, key, conf_val)
@@ -613,19 +610,19 @@ def _merged_profile(cls, values) -> Dict:
         values["merged_profile"] = merged_profile
         return values
 
-    @root_validator
-    def _validate_dynamo_no_retry(cls, values) -> Dict:
+    @model_validator(mode="after")
+    def _validate_dynamo_no_retry(self) -> Self:
         """Reject `retry` for services with a Dynamo router replica group.
         Dynamo workers cache the router's internal IP at provisioning time. A
         retry would produce a new router and likely a new internal_ip, leaving workers bound
         to a router that no longer exists.
         """
-        merged_profile = values.get("merged_profile")
-        cfg = values.get("configuration")
+        merged_profile = self.merged_profile
+        cfg = self.configuration
         if merged_profile is None or merged_profile.retry is None:
-            return values
+            return self
         if not isinstance(cfg, ServiceConfiguration):
-            return values
+            return self
         for g in cfg.replica_groups:
             if g.router is not None and g.router.type == RouterType.DYNAMO:
                 raise ValueError(
@@ -636,7 +633,7 @@ def _validate_dynamo_no_retry(cls, values) -> Dict:
                     "Remove `retry` from the profile/configuration and "
                     "re-apply."
                 )
-        return values
+        return self
 
 
 class ServiceModelSpec(CoreModel):
@@ -696,7 +693,7 @@ class Run(CoreModel):
     run_spec: RunSpec
     jobs: List[Job]
     latest_job_submission: Optional[JobSubmission] = None
-    cost: float = 0
+    cost: float = 0.0
     service: Optional[ServiceSpec] = None
     deployment_num: int = 0
     """`deployment_num` uses a default value for compatibility with pre-0.19.14 servers."""
diff --git a/src/dstack/_internal/core/models/volumes.py b/src/dstack/_internal/core/models/volumes.py
index f786804bfc..d651e717a4 100644
--- a/src/dstack/_internal/core/models/volumes.py
+++ b/src/dstack/_internal/core/models/volumes.py
@@ -4,13 +4,13 @@
 from pathlib import PurePosixPath
 from typing import Any, Dict, List, Literal, Optional, Tuple, Union
 
-from pydantic import Field, ValidationError, validator
+from pydantic import Field, RootModel, ValidationError, field_validator
 from typing_extensions import Annotated, Self
 
 from dstack._internal.core.errors import ConfigurationError
 from dstack._internal.core.models.backends.base import BackendType
 from dstack._internal.core.models.common import CoreModel
-from dstack._internal.core.models.profiles import parse_idle_duration
+from dstack._internal.core.models.duration import OptionalIdleDuration
 from dstack._internal.core.models.resources import Memory
 from dstack._internal.utils.common import get_or_error
 from dstack._internal.utils.tags import tags_validator
@@ -47,7 +47,7 @@ class BaseVolumeConfiguration(CoreModel):
         Field(description="The volume size. Must be specified when creating new volumes"),
     ] = None
     auto_cleanup_duration: Annotated[
-        Optional[Union[str, int]],
+        OptionalIdleDuration,
         Field(
             description=(
                 "Time to wait after volume is no longer used by any job before deleting it. "
@@ -67,10 +67,7 @@ class BaseVolumeConfiguration(CoreModel):
         ),
     ] = None
 
-    _validate_tags = validator("tags", pre=True, allow_reuse=True)(tags_validator)
-    _validate_auto_cleanup_duration = validator(
-        "auto_cleanup_duration", pre=True, allow_reuse=True
-    )(parse_idle_duration)
+    _validate_tags = field_validator("tags", mode="before")(tags_validator)
 
     @property
     def external_volume_id(self) -> Optional[str]:
@@ -184,13 +181,15 @@ def external_volume_id(self) -> Optional[str]:
 ]
 
 
-class VolumeConfiguration(CoreModel):
-    __root__: Annotated[AnyVolumeConfiguration, Field(discriminator="backend")]
+class VolumeConfiguration(
+    RootModel[Annotated[AnyVolumeConfiguration, Field(discriminator="backend")]]
+):
+    pass
 
 
 def parse_volume_configuration(data: dict) -> AnyVolumeConfiguration:
     try:
-        return VolumeConfiguration.parse_obj(data).__root__
+        return VolumeConfiguration.model_validate(data).root
     except ValidationError as e:
         raise ConfigurationError(e)
 
@@ -323,7 +322,7 @@ class VolumeMountPoint(CoreModel):
     ]
     path: Annotated[str, Field(description="The absolute container path to mount the volume at")]
 
-    _validate_path = validator("path", allow_reuse=True)(_validate_mount_point_path)
+    _validate_path = field_validator("path")(_validate_mount_point_path)
 
     @classmethod
     def parse(cls, v: str) -> Self:
@@ -344,10 +343,8 @@ class InstanceMountPoint(CoreModel):
         ),
     ] = False
 
-    _validate_instance_path = validator("instance_path", allow_reuse=True)(
-        _validate_mount_point_path
-    )
-    _validate_path = validator("path", allow_reuse=True)(_validate_mount_point_path)
+    _validate_instance_path = field_validator("instance_path")(_validate_mount_point_path)
+    _validate_path = field_validator("path")(_validate_mount_point_path)
 
     @classmethod
     def parse(cls, v: str) -> Self:
diff --git a/src/dstack/_internal/core/services/configs/__init__.py b/src/dstack/_internal/core/services/configs/__init__.py
index 8bc72a6394..4c88f6b061 100644
--- a/src/dstack/_internal/core/services/configs/__init__.py
+++ b/src/dstack/_internal/core/services/configs/__init__.py
@@ -28,13 +28,13 @@ def save(self):
         self.config_filepath.parent.mkdir(parents=True, exist_ok=True)
         with self.config_filepath.open("w") as f:
             # hack to convert enums to strings, etc.
-            yaml.dump(json.loads(self.config.json()), f)
+            yaml.dump(json.loads(self.config.model_dump_json()), f)
 
     def load(self):
         try:
             with open(self.config_filepath, "r") as f:
                 config = yaml.safe_load(f)
-            self.config = GlobalConfig.parse_obj(config)
+            self.config = GlobalConfig.model_validate(config)
         except FileNotFoundError:
             self.config = GlobalConfig()
         except ValidationError:
diff --git a/src/dstack/_internal/core/services/diff.py b/src/dstack/_internal/core/services/diff.py
index 154dee1db8..794d5c9ab9 100644
--- a/src/dstack/_internal/core/services/diff.py
+++ b/src/dstack/_internal/core/services/diff.py
@@ -33,14 +33,7 @@ def diff_models(
         A dict of changed fields in the form of
         `{<field_name>: {"old": old_value, "new": new_value}}`
     """
-    if not (
-        type(old) is type(new)
-        or (
-            isinstance(old, CoreModel)
-            and isinstance(new, CoreModel)
-            and type(old).__response__ is type(new).__response__
-        )
-    ):
+    if type(old) is not type(new):
         raise TypeError("Both instances must be of the same Pydantic model class.")
 
     if reset is not None:
@@ -48,7 +41,7 @@ def diff_models(
         new = copy_model(new, reset=reset)
 
     changes: ModelDiff = {}
-    for field in old.__fields__:
+    for field in type(old).model_fields:
         old_value = getattr(old, field)
         new_value = getattr(new, field)
         if old_value != new_value:
@@ -64,8 +57,8 @@ def copy_model(model: M, reset: Optional[IncludeExcludeType] = None) -> M:
     """
     Returns a deep copy of the model instance.
 
-    Implemented as `BaseModel.parse_obj(BaseModel.dict())`, thus,
-    unlike `BaseModel.copy(deep=True)`, runs all validations.
+    Implemented as `model_validate(model_dump())`, thus,
+    unlike `model_copy(deep=True)`, runs all validations.
 
     The fields specified in the `reset` option are reset to their default values.
 
@@ -75,7 +68,7 @@ def copy_model(model: M, reset: Optional[IncludeExcludeType] = None) -> M:
     Returns:
         A deep copy of the model instance.
     """
-    return type(model).parse_obj(model.dict(exclude=reset))
+    return type(model).model_validate(model.model_dump(exclude=reset))
 
 
 def flatten_diff_fields(diff: ModelDiff, prefix: str = "") -> list[str]:
diff --git a/src/dstack/_internal/core/services/profiles.py b/src/dstack/_internal/core/services/profiles.py
index 71ed2e520e..6850f45ba5 100644
--- a/src/dstack/_internal/core/services/profiles.py
+++ b/src/dstack/_internal/core/services/profiles.py
@@ -20,7 +20,7 @@ def get_retry(profile: Profile) -> Optional[Retry]:
                 duration=DEFAULT_RETRY_DURATION,
             )
         return None
-    profile_retry = profile_retry.copy()
+    profile_retry = profile_retry.model_copy()
     if profile_retry.on_events is None:
         profile_retry.on_events = [
             RetryEvent.NO_CAPACITY,
@@ -29,7 +29,7 @@ def get_retry(profile: Profile) -> Optional[Retry]:
         ]
     if profile_retry.duration is None:
         profile_retry.duration = DEFAULT_RETRY_DURATION
-    return Retry.parse_obj(profile_retry)
+    return Retry.model_validate(profile_retry.model_dump())
 
 
 def get_termination(
diff --git a/src/dstack/_internal/proxy/gateway/repo/repo.py b/src/dstack/_internal/proxy/gateway/repo/repo.py
index 0956faf04c..a3b69d4565 100644
--- a/src/dstack/_internal/proxy/gateway/repo/repo.py
+++ b/src/dstack/_internal/proxy/gateway/repo/repo.py
@@ -119,11 +119,11 @@ async def writer(self):
     @staticmethod
     def load(state_file: Path) -> "GatewayProxyRepo":
         if state_file.exists():
-            state = State.parse_file(state_file)
+            state = State.model_validate_json(state_file.read_text())
         else:
             state = None
         return GatewayProxyRepo(state=state, file=state_file)
 
     def save(self) -> None:
         if self._file is not None:
-            self._file.write_text(self._state.json())
+            self._file.write_text(self._state.model_dump_json())
diff --git a/src/dstack/_internal/proxy/gateway/services/model_routers/base.py b/src/dstack/_internal/proxy/gateway/services/model_routers/base.py
index 83ec14cb4d..a47677f400 100644
--- a/src/dstack/_internal/proxy/gateway/services/model_routers/base.py
+++ b/src/dstack/_internal/proxy/gateway/services/model_routers/base.py
@@ -2,7 +2,7 @@
 from pathlib import Path
 from typing import List, Literal, Optional
 
-from pydantic import BaseModel
+from pydantic import BaseModel, ConfigDict
 
 from dstack._internal.core.models.routers import AnyServiceRouterConfig
 
@@ -10,8 +10,7 @@
 class RouterContext(BaseModel):
     """Context for router initialization and configuration."""
 
-    class Config:
-        frozen = True
+    model_config = ConfigDict(frozen=True)
 
     host: str = "127.0.0.1"
     port: int
diff --git a/src/dstack/_internal/proxy/gateway/services/nginx.py b/src/dstack/_internal/proxy/gateway/services/nginx.py
index 9bc750b997..ca3700825a 100644
--- a/src/dstack/_internal/proxy/gateway/services/nginx.py
+++ b/src/dstack/_internal/proxy/gateway/services/nginx.py
@@ -37,7 +37,7 @@ class SiteConfig(BaseModel):
 
     def render(self) -> str:
         template = read_package_resource(f"{self.type}.jinja2")
-        render_dict = self.dict()
+        render_dict = self.model_dump()
         render_dict["proxy_port"] = PROXY_PORT_ON_GATEWAY
         return jinja2.Template(template).render(**render_dict)
 
diff --git a/src/dstack/_internal/proxy/gateway/services/registry.py b/src/dstack/_internal/proxy/gateway/services/registry.py
index f190523a39..3b7968d62b 100644
--- a/src/dstack/_internal/proxy/gateway/services/registry.py
+++ b/src/dstack/_internal/proxy/gateway/services/registry.py
@@ -443,7 +443,7 @@ async def _migrate_cors_enabled(repo: GatewayProxyRepo) -> None:
             not service.cors_enabled
             and (service.project_name, service.run_name) in openai_run_names
         ):
-            updated = models.Service(**{**service.dict(), "cors_enabled": True})
+            updated = models.Service(**{**service.model_dump(), "cors_enabled": True})
             await repo.set_service(updated)
 
 
diff --git a/src/dstack/_internal/proxy/lib/models.py b/src/dstack/_internal/proxy/lib/models.py
index a6128412d9..16e9fd4d6e 100644
--- a/src/dstack/_internal/proxy/lib/models.py
+++ b/src/dstack/_internal/proxy/lib/models.py
@@ -3,7 +3,7 @@
 from datetime import datetime
 from typing import Iterable, Literal, Optional, Union
 
-from pydantic import BaseModel, Field
+from pydantic import BaseModel, ConfigDict, Field
 from typing_extensions import Annotated
 
 from dstack._internal.core.models.instances import SSHConnectionParams
@@ -14,8 +14,7 @@
 # Models should be immutable so that they can be stored in memory and safely shared by
 # coroutines without copying on every read operation.
 class ImmutableModel(BaseModel):
-    class Config:
-        frozen = True
+    model_config = ConfigDict(frozen=True)
 
 
 class Replica(ImmutableModel):
@@ -38,11 +37,11 @@ class IPAddressPartitioningKey(ImmutableModel):
 
 class HeaderPartitioningKey(ImmutableModel):
     type: Literal["header"] = "header"
-    header: Annotated[str, Field(regex=r"^[a-zA-Z0-9-_]+$")]  # prevent Nginx config injection
+    header: Annotated[str, Field(pattern=r"^[a-zA-Z0-9-_]+$")]  # prevent Nginx config injection
 
 
 class RateLimit(ImmutableModel):
-    prefix: Annotated[str, Field(regex=r"^/[^\s\\{}]*$")]  # prevent Nginx config injection
+    prefix: Annotated[str, Field(pattern=r"^/[^\s\\{}]*$")]  # prevent Nginx config injection
     key: Annotated[
         Union[IPAddressPartitioningKey, HeaderPartitioningKey],
         Field(discriminator="type"),
@@ -78,7 +77,7 @@ def https_safe(self) -> bool:
         return self.https
 
     def with_replicas(self, new_replicas: Iterable[Replica]) -> "Service":
-        return Service(**{**self.dict(), "replicas": tuple(new_replicas)})
+        return Service(**{**self.model_dump(), "replicas": tuple(new_replicas)})
 
     def find_replica(self, replica_id: str) -> Optional[Replica]:
         for replica in self.replicas:
diff --git a/src/dstack/_internal/proxy/lib/routers/model_proxy.py b/src/dstack/_internal/proxy/lib/routers/model_proxy.py
index e5a5c4cee3..4e1cabab0a 100644
--- a/src/dstack/_internal/proxy/lib/routers/model_proxy.py
+++ b/src/dstack/_internal/proxy/lib/routers/model_proxy.py
@@ -99,4 +99,4 @@ async def _adaptor(self, first_chunk: Optional[ChatCompletionsChunk]) -> AsyncIt
 
     @staticmethod
     def _encode_chunk(chunk: ChatCompletionsChunk) -> bytes:
-        return f"data:{chunk.json()}\n\n".encode()
+        return f"data:{chunk.model_dump_json()}\n\n".encode()
diff --git a/src/dstack/_internal/proxy/lib/services/model_proxy/clients/openai.py b/src/dstack/_internal/proxy/lib/services/model_proxy/clients/openai.py
index ecd49823fd..dfd0e526f1 100644
--- a/src/dstack/_internal/proxy/lib/services/model_proxy/clients/openai.py
+++ b/src/dstack/_internal/proxy/lib/services/model_proxy/clients/openai.py
@@ -4,6 +4,7 @@
 from fastapi import status
 from pydantic import ValidationError
 
+from dstack._internal.core.models.common import validate_json_extra_ignore
 from dstack._internal.proxy.lib.errors import ProxyError
 from dstack._internal.proxy.lib.schemas.model_proxy import (
     ChatCompletionsChunk,
@@ -21,21 +22,23 @@ def __init__(self, http_client: httpx.AsyncClient, prefix: str):
     async def generate(self, request: ChatCompletionsRequest) -> ChatCompletionsResponse:
         try:
             resp = await self._http.post(
-                f"{self._prefix}/chat/completions", json=request.dict(exclude_unset=True)
+                f"{self._prefix}/chat/completions", json=request.model_dump(exclude_unset=True)
             )
             await self._propagate_error(resp)
         except httpx.RequestError as e:
             raise ProxyError(f"Error requesting model: {e!r}", status.HTTP_502_BAD_GATEWAY)
 
         try:
-            return ChatCompletionsResponse.__response__.parse_raw(resp.content)
+            return validate_json_extra_ignore(ChatCompletionsResponse, resp.content)
         except ValidationError as e:
             raise ProxyError(f"Invalid response from model: {e}", status.HTTP_502_BAD_GATEWAY)
 
     async def stream(self, request: ChatCompletionsRequest) -> AsyncIterator[ChatCompletionsChunk]:
         try:
             async with self._http.stream(
-                "POST", f"{self._prefix}/chat/completions", json=request.dict(exclude_unset=True)
+                "POST",
+                f"{self._prefix}/chat/completions",
+                json=request.model_dump(exclude_unset=True),
             ) as resp:
                 await self._propagate_error(resp)
 
@@ -52,7 +55,7 @@ async def stream(self, request: ChatCompletionsRequest) -> AsyncIterator[ChatCom
     @staticmethod
     def _parse_chunk_data(data: str) -> ChatCompletionsChunk:
         try:
-            return ChatCompletionsChunk.__response__.parse_raw(data)
+            return validate_json_extra_ignore(ChatCompletionsChunk, data)
         except ValidationError as e:
             raise ProxyError(f"Invalid chunk in model stream: {e}", status.HTTP_502_BAD_GATEWAY)
 
diff --git a/src/dstack/_internal/proxy/lib/services/model_proxy/clients/tgi.py b/src/dstack/_internal/proxy/lib/services/model_proxy/clients/tgi.py
index 70c8683a6c..2b1742acdc 100644
--- a/src/dstack/_internal/proxy/lib/services/model_proxy/clients/tgi.py
+++ b/src/dstack/_internal/proxy/lib/services/model_proxy/clients/tgi.py
@@ -79,7 +79,7 @@ async def generate(self, request: ChatCompletionsRequest) -> ChatCompletionsResp
         return ChatCompletionsResponse(
             id=uuid.uuid4().hex,
             choices=choices,
-            created=int(datetime.datetime.utcnow().timestamp()),
+            created=int(datetime.datetime.now(datetime.timezone.utc).timestamp()),
             model=request.model,
             system_fingerprint=f"fp_{data['details']['seed']}",
             usage=ChatCompletionsUsage(
@@ -91,7 +91,7 @@ async def generate(self, request: ChatCompletionsRequest) -> ChatCompletionsResp
 
     async def stream(self, request: ChatCompletionsRequest) -> AsyncIterator[ChatCompletionsChunk]:
         completion_id = uuid.uuid4().hex
-        created = int(datetime.datetime.utcnow().timestamp())
+        created = int(datetime.datetime.now(datetime.timezone.utc).timestamp())
 
         payload = self.get_payload(request)
         try:
diff --git a/src/dstack/_internal/server/app.py b/src/dstack/_internal/server/app.py
index bf884d001a..6380961e38 100644
--- a/src/dstack/_internal/server/app.py
+++ b/src/dstack/_internal/server/app.py
@@ -71,7 +71,7 @@
 from dstack._internal.server.utils import otel, sentry_utils
 from dstack._internal.server.utils.logging import configure_logging
 from dstack._internal.server.utils.routers import (
-    CustomORJSONResponse,
+    CustomJSONResponse,
     CustomStaticFiles,
     check_client_server_compatibility,
     error_detail,
@@ -270,14 +270,14 @@ async def forbidden_error_handler(request: Request, exc: ForbiddenError):
         msg = "Access denied"
         if len(exc.args) > 0:
             msg = exc.args[0]
-        return CustomORJSONResponse(
+        return CustomJSONResponse(
             status_code=status.HTTP_403_FORBIDDEN,
             content=error_detail(msg),
         )
 
     @app.exception_handler(ServerClientError)
     async def server_client_error_handler(request: Request, exc: ServerClientError):
-        return CustomORJSONResponse(
+        return CustomJSONResponse(
             status_code=status.HTTP_400_BAD_REQUEST,
             content={"detail": get_server_client_error_details(exc)},
         )
@@ -285,7 +285,7 @@ async def server_client_error_handler(request: Request, exc: ServerClientError):
     @app.exception_handler(OSError)
     async def os_error_handler(request, exc: OSError):
         if exc.errno in [36, 63]:
-            return CustomORJSONResponse(
+            return CustomJSONResponse(
                 {"detail": "Filename too long"},
                 status_code=status.HTTP_400_BAD_REQUEST,
             )
@@ -362,7 +362,7 @@ def _extract_endpoint_label(request: Request, response: Response) -> str:
 
     @app.get("/healthcheck")
     async def healthcheck():
-        return CustomORJSONResponse(content={"status": "running"})
+        return CustomJSONResponse(content={"status": "running"})
 
     if ui and Path(__file__).parent.joinpath("statics").exists():
         app.mount(
@@ -376,7 +376,7 @@ async def custom_http_exception_handler(request, exc):
                 or _is_proxy_request(request)
                 or _is_prometheus_request(request)
             ):
-                return CustomORJSONResponse(
+                return CustomJSONResponse(
                     {"detail": exc.detail},
                     status_code=status.HTTP_404_NOT_FOUND,
                 )
diff --git a/src/dstack/_internal/server/background/pipeline_tasks/instances/check.py b/src/dstack/_internal/server/background/pipeline_tasks/instances/check.py
index 779d83435a..9be54939c9 100644
--- a/src/dstack/_internal/server/background/pipeline_tasks/instances/check.py
+++ b/src/dstack/_internal/server/background/pipeline_tasks/instances/check.py
@@ -171,7 +171,7 @@ async def check_instance(instance_model: InstanceModel) -> ProcessResult:
             instance_id=instance_model.id,
             collected_at=get_current_datetime(),
             status=health_status,
-            response=instance_check.health_response.json(),
+            response=instance_check.health_response.model_dump_json(),
         )
 
     set_health_update(
@@ -357,7 +357,9 @@ async def _process_wait_for_instance_provisioning_data(
             instance_model.project.ssh_public_key,
             instance_model.project.ssh_private_key,
         )
-        result.instance_update_map["job_provisioning_data"] = job_provisioning_data.json()
+        result.instance_update_map["job_provisioning_data"] = (
+            job_provisioning_data.model_dump_json()
+        )
     except ProvisioningError as exc:
         logger.warning(
             "Error while waiting for instance %s to become running: %s",
diff --git a/src/dstack/_internal/server/background/pipeline_tasks/instances/cloud_provisioning.py b/src/dstack/_internal/server/background/pipeline_tasks/instances/cloud_provisioning.py
index e86862cd4e..cb87d0294a 100644
--- a/src/dstack/_internal/server/background/pipeline_tasks/instances/cloud_provisioning.py
+++ b/src/dstack/_internal/server/background/pipeline_tasks/instances/cloud_provisioning.py
@@ -213,9 +213,13 @@ async def create_cloud_instance(instance_model: InstanceModel) -> ProcessResult:
         result.instance_update_map["backend"] = backend.TYPE
         result.instance_update_map["region"] = instance_offer.region
         result.instance_update_map["price"] = instance_offer.price
-        result.instance_update_map["instance_configuration"] = instance_configuration.json()
-        result.instance_update_map["job_provisioning_data"] = job_provisioning_data.json()
-        result.instance_update_map["offer"] = instance_offer.json()
+        result.instance_update_map["instance_configuration"] = (
+            instance_configuration.model_dump_json()
+        )
+        result.instance_update_map["job_provisioning_data"] = (
+            job_provisioning_data.model_dump_json()
+        )
+        result.instance_update_map["offer"] = instance_offer.model_dump_json()
         result.instance_update_map["total_blocks"] = instance_offer.total_blocks
         result.instance_update_map["started_at"] = NOW_PLACEHOLDER
 
@@ -398,7 +402,7 @@ async def _find_or_create_suitable_placement_group_model(
             backend=instance_offer.backend,
             region=instance_offer.region,
             placement_strategy=PlacementStrategy.CLUSTER,
-        ).json(),
+        ).model_dump_json(),
     )
     placement_group = placement_group_model_to_placement_group(placement_group_model)
     logger.debug(
@@ -438,5 +442,5 @@ async def _find_or_create_suitable_placement_group_model(
         return None, False
 
     placement_group.provisioning_data = provisioning_data
-    placement_group_model.provisioning_data = provisioning_data.json()
+    placement_group_model.provisioning_data = provisioning_data.model_dump_json()
     return placement_group_model, True
diff --git a/src/dstack/_internal/server/background/pipeline_tasks/instances/common.py b/src/dstack/_internal/server/background/pipeline_tasks/instances/common.py
index 7ae5ad852a..ef12cebf98 100644
--- a/src/dstack/_internal/server/background/pipeline_tasks/instances/common.py
+++ b/src/dstack/_internal/server/background/pipeline_tasks/instances/common.py
@@ -193,8 +193,8 @@ def set_gpu_driver_update(
     if gpu_driver is None:
         return False
     current = job_provisioning_data.gpu_driver
-    if current is not None and current.dict() == gpu_driver.dict():
+    if current == gpu_driver:
         return False
     job_provisioning_data.gpu_driver = gpu_driver
-    update_map["job_provisioning_data"] = job_provisioning_data.json()
+    update_map["job_provisioning_data"] = job_provisioning_data.model_dump_json()
     return True
diff --git a/src/dstack/_internal/server/background/pipeline_tasks/instances/ssh_deploy.py b/src/dstack/_internal/server/background/pipeline_tasks/instances/ssh_deploy.py
index b4e3e1122a..c2c711cf24 100644
--- a/src/dstack/_internal/server/background/pipeline_tasks/instances/ssh_deploy.py
+++ b/src/dstack/_internal/server/background/pipeline_tasks/instances/ssh_deploy.py
@@ -17,6 +17,7 @@
 )
 from dstack._internal.core.errors import SSHProvisioningError
 from dstack._internal.core.models.backends.base import BackendType
+from dstack._internal.core.models.common import validate_json_extra_ignore
 from dstack._internal.core.models.instances import (
     InstanceAvailability,
     InstanceOfferWithAvailability,
@@ -52,7 +53,7 @@
     run_shim_as_systemd_service,
     upload_envs,
 )
-from dstack._internal.utils.common import get_current_datetime, run_async
+from dstack._internal.utils.common import get_current_datetime, get_or_error, run_async
 from dstack._internal.utils.logging import get_logger
 from dstack._internal.utils.network import get_ip_from_network, is_ip_among_addresses
 
@@ -194,8 +195,8 @@ async def add_ssh_instance(instance_model: InstanceModel) -> ProcessResult:
     )
     result.instance_update_map["backend"] = BackendType.REMOTE
     result.instance_update_map["price"] = 0
-    result.instance_update_map["offer"] = instance_offer.json()
-    result.instance_update_map["job_provisioning_data"] = job_provisioning_data.json()
+    result.instance_update_map["offer"] = instance_offer.model_dump_json()
+    result.instance_update_map["job_provisioning_data"] = job_provisioning_data.model_dump_json()
     result.instance_update_map["started_at"] = NOW_PLACEHOLDER
     result.instance_update_map["total_blocks"] = blocks
     return result
@@ -212,8 +213,8 @@ def _resolve_ssh_instance_network(
     instance_network = None
     internal_ip = None
     try:
-        default_job_provisioning_data = JobProvisioningData.__response__.parse_raw(
-            instance_model.job_provisioning_data
+        default_job_provisioning_data = validate_json_extra_ignore(
+            JobProvisioningData, get_or_error(instance_model.job_provisioning_data)
         )
         instance_network = default_job_provisioning_data.instance_network
         internal_ip = default_job_provisioning_data.internal_ip
@@ -295,7 +296,7 @@ def _deploy_instance(
 
         healthcheck_out = get_shim_healthcheck(client)
         try:
-            healthcheck = HealthcheckResponse.__response__.parse_raw(healthcheck_out)
+            healthcheck = validate_json_extra_ignore(HealthcheckResponse, healthcheck_out)
         except ValueError as exc:
             raise SSHProvisioningError(f"Cannot parse HealthcheckResponse: {exc}") from exc
         instance_check = runner_client.healthcheck_response_to_instance_check(healthcheck)
diff --git a/src/dstack/_internal/server/background/pipeline_tasks/jobs_running.py b/src/dstack/_internal/server/background/pipeline_tasks/jobs_running.py
index af65d4c485..720b0141ba 100644
--- a/src/dstack/_internal/server/background/pipeline_tasks/jobs_running.py
+++ b/src/dstack/_internal/server/background/pipeline_tasks/jobs_running.py
@@ -13,7 +13,11 @@
 
 from dstack._internal.core.consts import DSTACK_RUNNER_HTTP_PORT, DSTACK_SHIM_HTTP_PORT
 from dstack._internal.core.errors import GatewayError, SSHError
-from dstack._internal.core.models.common import NetworkMode, RegistryAuth
+from dstack._internal.core.models.common import (
+    NetworkMode,
+    RegistryAuth,
+    validate_json_extra_ignore,
+)
 from dstack._internal.core.models.configurations import (
     DevEnvironmentConfiguration,
     ServiceConfiguration,
@@ -413,7 +417,7 @@ async def _load_process_context(item: JobRunningPipelineItem) -> Optional[_Proce
             # gate in _prepare_startup_context can read its status / IP.
             # _fetch_run_model handles both: same-replica jobs always, plus
             # all non-terminated jobs when one exists.
-            run_spec = RunSpec.__response__.parse_raw(job_model.run.run_spec)
+            run_spec = validate_json_extra_ignore(RunSpec, job_model.run.run_spec)
             run_model = await _fetch_run_model(
                 session=session,
                 run_id=job_model.run_id,
@@ -872,7 +876,9 @@ async def _process_pulling_status(
             _set_job_runtime_data(result, shim_state.job_runtime_data)
 
         if shim_state.image_pull_progress is not None:
-            result.job_update_map["image_pull_progress"] = shim_state.image_pull_progress.json()
+            result.job_update_map["image_pull_progress"] = (
+                shim_state.image_pull_progress.model_dump_json()
+            )
 
         if shim_state.state == _ShimPullingState.WAITING:
             _reset_disconnected_at(context.job_model, result)
@@ -1215,7 +1221,7 @@ async def _register_service_replica(
     if context.run_model.gateway_id is None:
         return None
 
-    job_spec = JobSpec.__response__.parse_raw(context.job_model.job_spec_data)
+    job_spec = validate_json_extra_ignore(JobSpec, context.job_model.job_spec_data)
 
     # For router-based services (e.g. PD disaggregation), only router replicas should be
     # registered with the gateway. Worker replicas are discovered by the router-worker
@@ -1242,7 +1248,7 @@ async def _register_service_replica(
         instance_project_ssh_private_key = context.job_model.instance.project.ssh_private_key
     # JobRuntimeData might change on PULLING -> RUNNING path
     # so we must update job_submission with the result value.
-    job_submission = context.job_submission.copy(deep=True)
+    job_submission = context.job_submission.model_copy(deep=True)
     job_submission.job_runtime_data = _get_result_job_runtime_data(context.job_model, result)
     for conn in connections:
         try:
@@ -1391,7 +1397,7 @@ def _process_provisioning_with_shim(
     instance_mounts: list[InstanceMountPoint] = []
     for mount in run.run_spec.configuration.volumes:
         if isinstance(mount, VolumeMountPoint):
-            volume_mounts.append(mount.copy())
+            volume_mounts.append(mount.model_copy())
         elif isinstance(mount, InstanceMountPoint):
             instance_mounts.append(mount)
         else:
@@ -1513,7 +1519,7 @@ def _sync_shim_pulling_state(
                 task.termination_reason,
                 task.termination_message,
             )
-            logger.debug("task status: %s", task.dict())
+            logger.debug("task status: %s", task.model_dump())
             return _SyncShimPullingStateResult(
                 state=_ShimPullingState.FAILED,
                 termination_reason=JobTerminationReason(task.termination_reason.lower()),
@@ -1533,7 +1539,7 @@ def _sync_shim_pulling_state(
                     state=_ShimPullingState.WAITING,
                     image_pull_progress=image_pull_progress,
                 )
-            jrd = jrd.copy(update={"ports": {pm.container: pm.host for pm in task.ports}})
+            jrd = jrd.model_copy(update={"ports": {pm.container: pm.host for pm in task.ports}})
     else:
         shim_status = shim_client.pull()
         if (
@@ -1547,7 +1553,7 @@ def _sync_shim_pulling_state(
                 shim_status.result.reason,
                 shim_status.result.reason_message,
             )
-            logger.debug("shim status: %s", shim_status.dict())
+            logger.debug("shim status: %s", shim_status.model_dump())
             return _SyncShimPullingStateResult(
                 state=_ShimPullingState.FAILED,
                 termination_reason=JobTerminationReason(shim_status.result.reason.lower()),
@@ -1630,7 +1636,7 @@ def _submit_job_to_runner(
     job_info = runner_client.run_job()
     if job_info is not None:
         if jrd is not None:
-            jrd = jrd.copy(
+            jrd = jrd.model_copy(
                 update={"working_dir": job_info.working_dir, "username": job_info.username}
             )
     return _SubmitJobToRunnerResult(
@@ -1708,7 +1714,7 @@ def _terminate_if_inactivity_duration_exceeded(
     job_update_map: _JobUpdateMap,
     no_connections_secs: Optional[int],
 ) -> None:
-    conf = RunSpec.__response__.parse_raw(run_model.run_spec).configuration
+    conf = validate_json_extra_ignore(RunSpec, run_model.run_spec).configuration
     if not isinstance(conf, DevEnvironmentConfiguration) or not isinstance(
         conf.inactivity_duration, int
     ):
@@ -1896,7 +1902,7 @@ def _set_job_status(job_model: JobModel, result: _ProcessResult, new_status: Job
 
 
 def _set_job_runtime_data(result: _ProcessResult, jrd: Optional[JobRuntimeData]) -> None:
-    result.job_update_map["job_runtime_data"] = None if jrd is None else jrd.json()
+    result.job_update_map["job_runtime_data"] = None if jrd is None else jrd.model_dump_json()
 
 
 def _apply_submit_job_to_runner_result(
@@ -1929,7 +1935,7 @@ def _get_result_job_runtime_data(
     jrd = result.job_update_map.get("job_runtime_data", job_model.job_runtime_data)
     if jrd is None:
         return None
-    return JobRuntimeData.__response__.parse_raw(jrd)
+    return validate_json_extra_ignore(JobRuntimeData, jrd)
 
 
 def _get_result_registered(job_model: JobModel, result: _ProcessResult) -> bool:
diff --git a/src/dstack/_internal/server/background/pipeline_tasks/jobs_submitted.py b/src/dstack/_internal/server/background/pipeline_tasks/jobs_submitted.py
index 1824e0070e..74e1031c5f 100644
--- a/src/dstack/_internal/server/background/pipeline_tasks/jobs_submitted.py
+++ b/src/dstack/_internal/server/background/pipeline_tasks/jobs_submitted.py
@@ -21,7 +21,7 @@
     BACKENDS_WITH_PLACEMENT_GROUPS_SUPPORT,
 )
 from dstack._internal.core.errors import BackendError, ServerClientError, SkipOffer
-from dstack._internal.core.models.common import NetworkMode
+from dstack._internal.core.models.common import NetworkMode, validate_extra_ignore
 from dstack._internal.core.models.compute_groups import (
     ComputeGroupProvisioningData,
     ComputeGroupStatus,
@@ -891,8 +891,8 @@ def _get_master_job_provisioning_data(
     if master_job.job_submissions[-1].job_provisioning_data is None:
         return None
 
-    return JobProvisioningData.__response__.parse_obj(
-        master_job.job_submissions[-1].job_provisioning_data
+    return validate_extra_ignore(
+        JobProvisioningData, master_job.job_submissions[-1].job_provisioning_data
     )
 
 
@@ -1178,7 +1178,7 @@ def _assign_instance_to_job(
     job_model.instance = instance_model
     job_model.used_instance_id = instance_model.id
     job_model.job_provisioning_data = instance_model.job_provisioning_data
-    job_model.job_runtime_data = _prepare_job_runtime_data(offer, multinode).json()
+    job_model.job_runtime_data = _prepare_job_runtime_data(offer, multinode).model_dump_json()
     job_model.skip_min_processing_interval = True
 
     switch_instance_status(session, instance_model, InstanceStatus.BUSY)
@@ -1360,7 +1360,7 @@ async def _apply_existing_instance_provisioning(
         context.job_model.job_runtime_data = _prepare_job_runtime_data(
             offer=get_or_error(get_instance_offer(instance_model)),
             multinode=context.multinode,
-        ).json()
+        ).model_dump_json()
     switch_job_status(session, context.job_model, JobStatus.PROVISIONING)
     await _apply_volume_attachment_result(
         session=session,
@@ -1580,7 +1580,7 @@ def _resolve_provisioned_jobs_and_data(
             project=context.project,
             fleet=fleet_model,
             status=ComputeGroupStatus.RUNNING,
-            provisioning_data=provisioning_data.json(),
+            provisioning_data=provisioning_data.model_dump_json(),
         )
         return (
             context.jobs_to_provision,
@@ -1611,7 +1611,7 @@ async def _promote_or_create_instance_models_for_provisioned_jobs(
         provisioned_job_models, job_provisioning_datas
     ):
         provisioned_job_model.fleet_id = fleet_model.id
-        provisioned_job_model.job_provisioning_data = job_provisioning_data.json()
+        provisioned_job_model.job_provisioning_data = job_provisioning_data.model_dump_json()
         switch_job_status(session, provisioned_job_model, JobStatus.PROVISIONING)
         provisioned_job_model.skip_min_processing_interval = True
 
@@ -1646,7 +1646,7 @@ async def _promote_or_create_instance_models_for_provisioned_jobs(
         instance_models.append(instance_model)
         provisioned_job_model.job_runtime_data = _prepare_job_runtime_data(
             offer, context.multinode
-        ).json()
+        ).model_dump_json()
         events.emit(
             session,
             f"Instance provisioned for job. Instance status: {instance_model.status.upper()}",
@@ -1714,8 +1714,8 @@ def _create_instance_model_for_job(
         started_at=get_current_datetime(),
         status=InstanceStatus.PROVISIONING,
         unreachable=False,
-        job_provisioning_data=job_provisioning_data.json(),
-        offer=offer.json(),
+        job_provisioning_data=job_provisioning_data.model_dump_json(),
+        offer=offer.model_dump_json(),
         termination_policy=termination_policy,
         termination_idle_time=termination_idle_time,
         jobs=[job_model],
@@ -1747,8 +1747,8 @@ def _promote_placeholder_instance(
     instance_model.status = InstanceStatus.PROVISIONING
     instance_model.started_at = get_current_datetime()
     instance_model.compute_group = compute_group_model
-    instance_model.job_provisioning_data = job_provisioning_data.json()
-    instance_model.offer = offer.json()
+    instance_model.job_provisioning_data = job_provisioning_data.model_dump_json()
+    instance_model.offer = offer.model_dump_json()
     instance_model.backend = offer.backend
     instance_model.price = offer.price
     instance_model.region = offer.region
@@ -1822,7 +1822,7 @@ async def _process_volume_attachments(
                 attachments.append(
                     _VolumeAttachmentPayload(
                         volume_id=volume_model.id,
-                        attachment_data=attachment_data.json(),
+                        attachment_data=attachment_data.model_dump_json(),
                         volume_name=volume.name,
                     )
                 )
@@ -1990,7 +1990,7 @@ async def _apply_volume_attachment_result(
     job_runtime_data.volume_names = [
         attachment.volume_name for attachment in volume_attachment_result.attachments
     ]
-    job_model.job_runtime_data = job_runtime_data.json()
+    job_model.job_runtime_data = job_runtime_data.model_dump_json()
 
     volume_ids = [attachment.volume_id for attachment in volume_attachment_result.attachments]
     if volume_ids:
diff --git a/src/dstack/_internal/server/background/pipeline_tasks/volumes.py b/src/dstack/_internal/server/background/pipeline_tasks/volumes.py
index 66dabdeb07..a5d08ac803 100644
--- a/src/dstack/_internal/server/background/pipeline_tasks/volumes.py
+++ b/src/dstack/_internal/server/background/pipeline_tasks/volumes.py
@@ -366,7 +366,7 @@ async def _process_submitted_volume(volume_model: VolumeModel) -> _ProcessResult
     return _ProcessResult(
         update_map={
             "status": VolumeStatus.ACTIVE,
-            "volume_provisioning_data": vpd.json(),
+            "volume_provisioning_data": vpd.model_dump_json(),
         }
     )
 
diff --git a/src/dstack/_internal/server/background/scheduled_tasks/idle_volumes.py b/src/dstack/_internal/server/background/scheduled_tasks/idle_volumes.py
index e1625131e1..cf41b977b1 100644
--- a/src/dstack/_internal/server/background/scheduled_tasks/idle_volumes.py
+++ b/src/dstack/_internal/server/background/scheduled_tasks/idle_volumes.py
@@ -5,7 +5,7 @@
 from sqlalchemy.ext.asyncio import AsyncSession
 from sqlalchemy.orm import joinedload
 
-from dstack._internal.core.models.profiles import parse_duration
+from dstack._internal.core.models.duration import parse_duration
 from dstack._internal.core.models.volumes import VolumeStatus
 from dstack._internal.server.db import get_db, get_session_ctx
 from dstack._internal.server.models import ProjectModel, UserModel, VolumeModel
diff --git a/src/dstack/_internal/server/models.py b/src/dstack/_internal/server/models.py
index a70ea9e5e6..72c9745422 100644
--- a/src/dstack/_internal/server/models.py
+++ b/src/dstack/_internal/server/models.py
@@ -3,6 +3,7 @@
 from datetime import datetime, timezone
 from typing import Callable, Generic, List, Optional, TypeVar, Union
 
+from pydantic import ConfigDict
 from sqlalchemy import (
     BigInteger,
     Boolean,
@@ -24,7 +25,7 @@
 
 from dstack._internal.core.errors import DstackError
 from dstack._internal.core.models.backends.base import BackendType
-from dstack._internal.core.models.common import CoreConfig, generate_dual_core_model
+from dstack._internal.core.models.common import CoreModel
 from dstack._internal.core.models.compute_groups import ComputeGroupStatus
 from dstack._internal.core.models.events import EventTargetType
 from dstack._internal.core.models.fleets import FleetStatus
@@ -76,11 +77,9 @@ def process_result_value(self, value, dialect):
         return value.replace(tzinfo=timezone.utc)
 
 
-class DecryptedStringConfig(CoreConfig):
-    arbitrary_types_allowed = True
+class DecryptedString(CoreModel):
+    model_config = ConfigDict(arbitrary_types_allowed=True)
 
-
-class DecryptedString(generate_dual_core_model(DecryptedStringConfig)):
     """
     A type for representing plaintext strings encrypted with `EncryptedString`.
     Besides the string, stores information if the decryption was successful.
diff --git a/src/dstack/_internal/server/routers/auth.py b/src/dstack/_internal/server/routers/auth.py
index e44fb67f53..042739f89c 100644
--- a/src/dstack/_internal/server/routers/auth.py
+++ b/src/dstack/_internal/server/routers/auth.py
@@ -6,7 +6,7 @@
     OAuthGetNextRedirectResponse,
 )
 from dstack._internal.server.services import auth as auth_services
-from dstack._internal.server.utils.routers import CustomORJSONResponse
+from dstack._internal.server.utils.routers import CustomJSONResponse
 
 router = APIRouter(prefix="/api/auth", tags=["authentication"])
 
@@ -18,7 +18,7 @@ async def list_providers():
     """
     Returns OAuth2 providers registered on the server.
     """
-    return CustomORJSONResponse(auth_services.list_providers())
+    return CustomJSONResponse(auth_services.list_providers())
 
 
 @router.post(
@@ -33,7 +33,7 @@ async def get_next_redirect(body: OAuthGetNextRedirectRequest):
     to determine if the user needs to be redirected further (CLI login)
     or the auth callback endpoint needs to be called directly (UI login).
     """
-    return CustomORJSONResponse(
+    return CustomJSONResponse(
         OAuthGetNextRedirectResponse(
             redirect_url=auth_services.get_next_redirect_url(code=body.code, state=body.state)
         )
diff --git a/src/dstack/_internal/server/routers/backends.py b/src/dstack/_internal/server/routers/backends.py
index 4fd84a4477..8c79993269 100644
--- a/src/dstack/_internal/server/routers/backends.py
+++ b/src/dstack/_internal/server/routers/backends.py
@@ -28,7 +28,7 @@
     update_backend_config_yaml,
 )
 from dstack._internal.server.utils.routers import (
-    CustomORJSONResponse,
+    CustomJSONResponse,
     get_base_api_additional_responses,
 )
 
@@ -46,7 +46,7 @@
 
 @root_router.post("/list_types", summary="List backend types", response_model=List[BackendType])
 async def list_backend_types():
-    return CustomORJSONResponse(
+    return CustomJSONResponse(
         dstack._internal.core.backends.configurators.list_available_backend_types()
     )
 
@@ -61,7 +61,7 @@ async def create_backend(
     config = await backends.create_backend(session=session, project=project, config=body)
     if settings.SERVER_CONFIG_ENABLED:
         await ServerConfigManager().sync_config(session=session)
-    return CustomORJSONResponse(config)
+    return CustomJSONResponse(config)
 
 
 @project_router.post("/update", summary="Update backend", response_model=AnyBackendConfigWithCreds)
@@ -74,7 +74,7 @@ async def update_backend(
     config = await backends.update_backend(session=session, project=project, config=body)
     if settings.SERVER_CONFIG_ENABLED:
         await ServerConfigManager().sync_config(session=session)
-    return CustomORJSONResponse(config)
+    return CustomJSONResponse(config)
 
 
 @project_router.post("/delete", summary="Delete backends")
@@ -104,7 +104,7 @@ async def get_backend_config_info(
     config = await backends.get_backend_config(project=project, backend_type=backend_name)
     if config is None:
         raise ResourceNotExistsError()
-    return CustomORJSONResponse(config)
+    return CustomJSONResponse(config)
 
 
 @project_router.post("/create_yaml", summary="Create backend YAML")
@@ -143,6 +143,6 @@ async def get_backend_yaml(
     user_project: Tuple[UserModel, ProjectModel] = Depends(ProjectAdmin()),
 ):
     _, project = user_project
-    return CustomORJSONResponse(
+    return CustomJSONResponse(
         await get_backend_config_yaml(project=project, backend_type=backend_name)
     )
diff --git a/src/dstack/_internal/server/routers/events.py b/src/dstack/_internal/server/routers/events.py
index 1d6d80b671..574a59bac8 100644
--- a/src/dstack/_internal/server/routers/events.py
+++ b/src/dstack/_internal/server/routers/events.py
@@ -8,7 +8,7 @@
 from dstack._internal.server.schemas.events import ListEventsRequest
 from dstack._internal.server.security.permissions import Authenticated
 from dstack._internal.server.utils.routers import (
-    CustomORJSONResponse,
+    CustomJSONResponse,
     get_base_api_additional_responses,
 )
 
@@ -38,7 +38,7 @@ async def list_events(
     This should be taken into account when using the API to monitor recent events,
     so that delayed events are not missed during pagination.
     """
-    return CustomORJSONResponse(
+    return CustomJSONResponse(
         await events_services.list_events(
             session=session,
             user=user,
diff --git a/src/dstack/_internal/server/routers/files.py b/src/dstack/_internal/server/routers/files.py
index 5456d3d5d5..c255e600a9 100644
--- a/src/dstack/_internal/server/routers/files.py
+++ b/src/dstack/_internal/server/routers/files.py
@@ -12,7 +12,7 @@
 from dstack._internal.server.services import files
 from dstack._internal.server.settings import SERVER_CODE_UPLOAD_LIMIT
 from dstack._internal.server.utils.routers import (
-    CustomORJSONResponse,
+    CustomJSONResponse,
     get_base_api_additional_responses,
     get_request_size,
 )
@@ -40,7 +40,7 @@ async def get_archive_by_hash(
     )
     if archive is None:
         raise ResourceNotExistsError()
-    return CustomORJSONResponse(archive)
+    return CustomJSONResponse(archive)
 
 
 @router.post("/upload_archive", summary="Upload file archive", response_model=FileArchive)
@@ -67,4 +67,4 @@ async def upload_archive(
         user=user,
         file=file,
     )
-    return CustomORJSONResponse(archive)
+    return CustomJSONResponse(archive)
diff --git a/src/dstack/_internal/server/routers/fleets.py b/src/dstack/_internal/server/routers/fleets.py
index b4131ee022..5abf7ce19a 100644
--- a/src/dstack/_internal/server/routers/fleets.py
+++ b/src/dstack/_internal/server/routers/fleets.py
@@ -28,7 +28,7 @@
 )
 from dstack._internal.server.services.pipelines import PipelineHinterProtocol, get_pipeline_hinter
 from dstack._internal.server.utils.routers import (
-    CustomORJSONResponse,
+    CustomJSONResponse,
     get_base_api_additional_responses,
     get_client_version,
 )
@@ -73,7 +73,7 @@ async def list_fleets(
     )
     for fleet in fleet_list:
         patch_fleet(fleet, client_version)
-    return CustomORJSONResponse(fleet_list)
+    return CustomJSONResponse(fleet_list)
 
 
 @project_router.post("/list", summary="List project fleets", response_model=List[Fleet])
@@ -97,7 +97,7 @@ async def list_project_fleets(
     )
     for fleet in fleet_list:
         patch_fleet(fleet, client_version)
-    return CustomORJSONResponse(fleet_list)
+    return CustomJSONResponse(fleet_list)
 
 
 @project_router.post("/get", summary="Get fleet", response_model=Fleet)
@@ -122,7 +122,7 @@ async def get_fleet(
     if fleet is None:
         raise ResourceNotExistsError()
     patch_fleet(fleet, client_version)
-    return CustomORJSONResponse(fleet)
+    return CustomJSONResponse(fleet)
 
 
 @project_router.post("/get_plan", summary="Get fleet plan", response_model=FleetPlan)
@@ -143,7 +143,7 @@ async def get_plan(
         spec=body.spec,
     )
     patch_fleet_plan(plan, client_version)
-    return CustomORJSONResponse(plan)
+    return CustomJSONResponse(plan)
 
 
 @project_router.post("/apply", summary="Apply fleet plan", response_model=Fleet)
@@ -169,7 +169,7 @@ async def apply_plan(
         pipeline_hinter=pipeline_hinter,
     )
     patch_fleet(fleet, client_version)
-    return CustomORJSONResponse(fleet)
+    return CustomJSONResponse(fleet)
 
 
 @project_router.post("/create", summary="Create fleet", response_model=Fleet, deprecated=True)
@@ -192,7 +192,7 @@ async def create_fleet(
         pipeline_hinter=pipeline_hinter,
     )
     patch_fleet(fleet, client_version)
-    return CustomORJSONResponse(fleet)
+    return CustomJSONResponse(fleet)
 
 
 @project_router.post("/delete", summary="Delete fleets")
diff --git a/src/dstack/_internal/server/routers/gateways.py b/src/dstack/_internal/server/routers/gateways.py
index 134ce04eae..26764a81b9 100644
--- a/src/dstack/_internal/server/routers/gateways.py
+++ b/src/dstack/_internal/server/routers/gateways.py
@@ -21,7 +21,7 @@
 )
 from dstack._internal.server.services.pipelines import PipelineHinterProtocol, get_pipeline_hinter
 from dstack._internal.server.utils.routers import (
-    CustomORJSONResponse,
+    CustomJSONResponse,
     get_base_api_additional_responses,
     get_client_version,
 )
@@ -50,7 +50,7 @@ async def list_gateways(
     )
     for gateway in gateway_list:
         patch_gateway(gateway, client_version)
-    return CustomORJSONResponse(gateway_list)
+    return CustomJSONResponse(gateway_list)
 
 
 @router.post("/get", summary="Get gateway", response_model=models.Gateway)
@@ -68,7 +68,7 @@ async def get_gateway(
     if gateway is None:
         raise ResourceNotExistsError()
     patch_gateway(gateway, client_version)
-    return CustomORJSONResponse(gateway)
+    return CustomJSONResponse(gateway)
 
 
 @router.post("/get_plan", summary="Get gateway plan", response_model=models.GatewayPlan)
@@ -90,7 +90,7 @@ async def get_plan(
         spec=body.spec,
     )
     patch_gateway_plan(plan, client_version)
-    return CustomORJSONResponse(plan)
+    return CustomJSONResponse(plan)
 
 
 @router.post("/apply", summary="Apply gateway plan", response_model=models.Gateway)
@@ -114,7 +114,7 @@ async def apply_plan(
         pipeline_hinter=pipeline_hinter,
     )
     patch_gateway(gateway, client_version)
-    return CustomORJSONResponse(gateway)
+    return CustomJSONResponse(gateway)
 
 
 @router.post("/create", summary="Create gateway", response_model=models.Gateway, deprecated=True)
@@ -137,7 +137,7 @@ async def create_gateway(
         pipeline_hinter=pipeline_hinter,
     )
     patch_gateway(gateway, client_version)
-    return CustomORJSONResponse(gateway)
+    return CustomJSONResponse(gateway)
 
 
 @router.post("/delete", summary="Delete gateways")
@@ -194,4 +194,4 @@ async def set_gateway_wildcard_domain(
         user=user,
     )
     patch_gateway(gateway, client_version)
-    return CustomORJSONResponse(gateway)
+    return CustomJSONResponse(gateway)
diff --git a/src/dstack/_internal/server/routers/instances.py b/src/dstack/_internal/server/routers/instances.py
index 4b96d422c2..ebe3b8c390 100644
--- a/src/dstack/_internal/server/routers/instances.py
+++ b/src/dstack/_internal/server/routers/instances.py
@@ -21,7 +21,7 @@
     check_can_access_instance,
 )
 from dstack._internal.server.utils.routers import (
-    CustomORJSONResponse,
+    CustomJSONResponse,
     get_base_api_additional_responses,
 )
 
@@ -50,7 +50,7 @@ async def list_instances(
     The results are paginated. To get the next page, pass `created_at` and `id` of
     the last instance from the previous page as `prev_created_at` and `prev_id`.
     """
-    return CustomORJSONResponse(
+    return CustomJSONResponse(
         await instances_services.list_user_instances(
             session=session,
             user=user,
@@ -82,7 +82,7 @@ async def get_instance_health_checks(
         before=body.before,
         limit=body.limit,
     )
-    return CustomORJSONResponse(GetInstanceHealthChecksResponse(health_checks=health_checks))
+    return CustomJSONResponse(GetInstanceHealthChecksResponse(health_checks=health_checks))
 
 
 @project_router.post("/get", response_model=Instance)
@@ -103,4 +103,4 @@ async def get_instance(
     )
     if instance is None:
         raise ResourceNotExistsError()
-    return CustomORJSONResponse(instance)
+    return CustomJSONResponse(instance)
diff --git a/src/dstack/_internal/server/routers/logs.py b/src/dstack/_internal/server/routers/logs.py
index 540c4ec125..fa51787133 100644
--- a/src/dstack/_internal/server/routers/logs.py
+++ b/src/dstack/_internal/server/routers/logs.py
@@ -8,7 +8,7 @@
 from dstack._internal.server.security.permissions import ProjectMember
 from dstack._internal.server.services import logs
 from dstack._internal.server.utils.routers import (
-    CustomORJSONResponse,
+    CustomJSONResponse,
     get_base_api_additional_responses,
 )
 
@@ -32,4 +32,4 @@ async def poll_logs(
     # The runner guarantees logs have different timestamps if throughput < 1k logs / sec.
     # Otherwise, some logs with duplicated timestamps may be filtered out.
     # This limitation is imposed by cloud log services that support up to millisecond timestamp resolution.
-    return CustomORJSONResponse(await logs.poll_logs_async(project=project, request=body))
+    return CustomJSONResponse(await logs.poll_logs_async(project=project, request=body))
diff --git a/src/dstack/_internal/server/routers/metrics.py b/src/dstack/_internal/server/routers/metrics.py
index 14d0eb6fb5..b44d4a86a3 100644
--- a/src/dstack/_internal/server/routers/metrics.py
+++ b/src/dstack/_internal/server/routers/metrics.py
@@ -13,7 +13,7 @@
 from dstack._internal.server.services import metrics
 from dstack._internal.server.services.jobs import get_run_job_model
 from dstack._internal.server.utils.routers import (
-    CustomORJSONResponse,
+    CustomJSONResponse,
     get_base_api_additional_responses,
 )
 
@@ -72,7 +72,7 @@ async def get_job_metrics(
     if job_model is None:
         raise ResourceNotExistsError("Found no job with given parameters")
 
-    return CustomORJSONResponse(
+    return CustomJSONResponse(
         await metrics.get_job_metrics(
             session=session,
             job_model=job_model,
diff --git a/src/dstack/_internal/server/routers/projects.py b/src/dstack/_internal/server/routers/projects.py
index f7d9098dfc..a3b8f8ba2f 100644
--- a/src/dstack/_internal/server/routers/projects.py
+++ b/src/dstack/_internal/server/routers/projects.py
@@ -25,7 +25,7 @@
 )
 from dstack._internal.server.services import fleets, projects
 from dstack._internal.server.utils.routers import (
-    CustomORJSONResponse,
+    CustomJSONResponse,
     get_base_api_additional_responses,
 )
 
@@ -53,7 +53,7 @@ async def list_projects(
     if body is None:
         # For backward compatibility
         body = ListProjectsRequest()
-    return CustomORJSONResponse(
+    return CustomJSONResponse(
         await projects.list_user_accessible_projects(
             session=session,
             user=user,
@@ -86,7 +86,7 @@ async def list_only_no_fleets(
 
     `members` and `backends` are always empty - call `/api/projects/{project_name}/get` to retrieve them.
     """
-    return CustomORJSONResponse(
+    return CustomJSONResponse(
         await fleets.list_projects_with_no_active_fleets(session=session, user=user)
     )
 
@@ -97,7 +97,7 @@ async def create_project(
     session: AsyncSession = Depends(get_session),
     user: UserModel = Depends(Authenticated()),
 ):
-    return CustomORJSONResponse(
+    return CustomJSONResponse(
         await projects.create_project(
             session=session,
             user=user,
@@ -127,7 +127,7 @@ async def get_project(
     user_project: Tuple[UserModel, ProjectModel] = Depends(ProjectMemberOrPublicAccess()),
 ):
     _, project = user_project
-    return CustomORJSONResponse(projects.project_model_to_project(project))
+    return CustomJSONResponse(projects.project_model_to_project(project))
 
 
 @router.post(
@@ -148,7 +148,7 @@ async def set_project_members(
         members=body.members,
     )
     await session.refresh(project)
-    return CustomORJSONResponse(projects.project_model_to_project(project))
+    return CustomJSONResponse(projects.project_model_to_project(project))
 
 
 @router.post(
@@ -169,7 +169,7 @@ async def add_project_members(
         members=body.members,
     )
     await session.refresh(project)
-    return CustomORJSONResponse(projects.project_model_to_project(project))
+    return CustomJSONResponse(projects.project_model_to_project(project))
 
 
 @router.post(
@@ -190,7 +190,7 @@ async def remove_project_members(
         usernames=body.usernames,
     )
     await session.refresh(project)
-    return CustomORJSONResponse(projects.project_model_to_project(project))
+    return CustomJSONResponse(projects.project_model_to_project(project))
 
 
 @router.post(
@@ -213,4 +213,4 @@ async def update_project(
         reset_templates_repo=body.reset_templates_repo,
     )
     await session.refresh(project)
-    return CustomORJSONResponse(projects.project_model_to_project(project))
+    return CustomJSONResponse(projects.project_model_to_project(project))
diff --git a/src/dstack/_internal/server/routers/public_keys.py b/src/dstack/_internal/server/routers/public_keys.py
index e846d15415..998d38bc00 100644
--- a/src/dstack/_internal/server/routers/public_keys.py
+++ b/src/dstack/_internal/server/routers/public_keys.py
@@ -13,7 +13,7 @@
 from dstack._internal.server.security.permissions import Authenticated
 from dstack._internal.server.services import public_keys as public_keys_services
 from dstack._internal.server.utils.routers import (
-    CustomORJSONResponse,
+    CustomJSONResponse,
     get_base_api_additional_responses,
 )
 
@@ -30,7 +30,7 @@ async def list_user_public_keys(
     user: Annotated[UserModel, Depends(Authenticated())],
 ):
     public_keys = await public_keys_services.list_user_public_keys(session=session, user=user)
-    return CustomORJSONResponse(public_keys)
+    return CustomJSONResponse(public_keys)
 
 
 @router.post("/add", summary="Add SSH key", response_model=PublicKeyInfo)
@@ -42,7 +42,7 @@ async def add_user_public_key(
     public_key = await public_keys_services.add_user_public_key(
         session=session, user=user, key=body.key, name=body.name
     )
-    return CustomORJSONResponse(public_key)
+    return CustomJSONResponse(public_key)
 
 
 @router.post("/delete", summary="Delete SSH keys")
diff --git a/src/dstack/_internal/server/routers/repos.py b/src/dstack/_internal/server/routers/repos.py
index 95307bd15c..c45063d1b7 100644
--- a/src/dstack/_internal/server/routers/repos.py
+++ b/src/dstack/_internal/server/routers/repos.py
@@ -16,7 +16,7 @@
 from dstack._internal.server.services import repos
 from dstack._internal.server.settings import SERVER_CODE_UPLOAD_LIMIT
 from dstack._internal.server.utils.routers import (
-    CustomORJSONResponse,
+    CustomJSONResponse,
     get_base_api_additional_responses,
     get_request_size,
 )
@@ -35,7 +35,7 @@ async def list_repos(
     user_project: Tuple[UserModel, ProjectModel] = Depends(ProjectMember()),
 ):
     _, project = user_project
-    return CustomORJSONResponse(await repos.list_repos(session=session, project=project))
+    return CustomJSONResponse(await repos.list_repos(session=session, project=project))
 
 
 @router.post("/get", summary="Get repo", response_model=RepoHeadWithCreds)
@@ -54,7 +54,7 @@ async def get_repo(
     )
     if repo is None:
         raise ResourceNotExistsError()
-    return CustomORJSONResponse(repo)
+    return CustomJSONResponse(repo)
 
 
 @router.post("/init", summary="Initialize repo")
diff --git a/src/dstack/_internal/server/routers/runs.py b/src/dstack/_internal/server/routers/runs.py
index 994b91fa29..c2434e1469 100644
--- a/src/dstack/_internal/server/routers/runs.py
+++ b/src/dstack/_internal/server/routers/runs.py
@@ -23,7 +23,7 @@
 from dstack._internal.server.services import runs, users
 from dstack._internal.server.services.pipelines import PipelineHinterProtocol, get_pipeline_hinter
 from dstack._internal.server.utils.routers import (
-    CustomORJSONResponse,
+    CustomJSONResponse,
     get_base_api_additional_responses,
     get_client_version,
 )
@@ -86,7 +86,7 @@ async def list_runs(
     )
     for run in run_list:
         patch_run(run, client_version)
-    return CustomORJSONResponse(run_list)
+    return CustomJSONResponse(run_list)
 
 
 @project_router.post("/get", response_model=Run, summary="Get run")
@@ -111,7 +111,7 @@ async def get_run(
     if run is None:
         raise ResourceNotExistsError("Run not found")
     patch_run(run, client_version)
-    return CustomORJSONResponse(run)
+    return CustomJSONResponse(run)
 
 
 @project_router.post(
@@ -144,7 +144,7 @@ async def get_plan(
         legacy_repo_dir=legacy_repo_dir,
     )
     patch_run_plan(run_plan, client_version)
-    return CustomORJSONResponse(run_plan)
+    return CustomJSONResponse(run_plan)
 
 
 @project_router.post("/apply", response_model=Run, summary="Apply run plan")
@@ -175,7 +175,7 @@ async def apply_plan(
         legacy_repo_dir=legacy_repo_dir,
     )
     patch_run(run, client_version)
-    return CustomORJSONResponse(run)
+    return CustomJSONResponse(run)
 
 
 @project_router.post("/stop", summary="Stop runs")
diff --git a/src/dstack/_internal/server/routers/secrets.py b/src/dstack/_internal/server/routers/secrets.py
index db4d364b36..c1b8ec0f79 100644
--- a/src/dstack/_internal/server/routers/secrets.py
+++ b/src/dstack/_internal/server/routers/secrets.py
@@ -14,7 +14,7 @@
 )
 from dstack._internal.server.security.permissions import ProjectManager
 from dstack._internal.server.services import secrets as secrets_services
-from dstack._internal.server.utils.routers import CustomORJSONResponse
+from dstack._internal.server.utils.routers import CustomJSONResponse
 
 router = APIRouter(
     prefix="/api/project/{project_name}/secrets",
@@ -28,7 +28,7 @@ async def list_secrets(
     user_project: Tuple[UserModel, ProjectModel] = Depends(ProjectManager()),
 ):
     user, project = user_project
-    return CustomORJSONResponse(
+    return CustomJSONResponse(
         await secrets_services.list_secrets(
             session=session,
             project=project,
@@ -52,7 +52,7 @@ async def get_secret(
     )
     if secret is None:
         raise ResourceNotExistsError()
-    return CustomORJSONResponse(secret)
+    return CustomJSONResponse(secret)
 
 
 @router.post("/create_or_update", summary="Create or update secret", response_model=Secret)
@@ -62,7 +62,7 @@ async def create_or_update_secret(
     user_project: Tuple[UserModel, ProjectModel] = Depends(ProjectManager()),
 ):
     user, project = user_project
-    return CustomORJSONResponse(
+    return CustomJSONResponse(
         await secrets_services.create_or_update_secret(
             session=session,
             project=project,
diff --git a/src/dstack/_internal/server/routers/server.py b/src/dstack/_internal/server/routers/server.py
index 8fd3d77d46..2d2e7123fb 100644
--- a/src/dstack/_internal/server/routers/server.py
+++ b/src/dstack/_internal/server/routers/server.py
@@ -2,7 +2,7 @@
 
 from dstack._internal import settings
 from dstack._internal.core.models.server import ServerInfo
-from dstack._internal.server.utils.routers import CustomORJSONResponse
+from dstack._internal.server.utils.routers import CustomJSONResponse
 
 router = APIRouter(
     prefix="/api/server",
@@ -12,7 +12,7 @@
 
 @router.post("/get_info", summary="Get server info", response_model=ServerInfo)
 async def get_server_info():
-    return CustomORJSONResponse(
+    return CustomJSONResponse(
         ServerInfo(
             server_version=settings.DSTACK_VERSION,
         )
diff --git a/src/dstack/_internal/server/routers/sshproxy.py b/src/dstack/_internal/server/routers/sshproxy.py
index 0baeb0f0ed..3c472fbc73 100644
--- a/src/dstack/_internal/server/routers/sshproxy.py
+++ b/src/dstack/_internal/server/routers/sshproxy.py
@@ -10,7 +10,7 @@
 from dstack._internal.server.security.permissions import AlwaysForbidden, ServiceAccount
 from dstack._internal.server.services.sshproxy.handlers import get_upstream_response
 from dstack._internal.server.utils.routers import (
-    CustomORJSONResponse,
+    CustomJSONResponse,
     get_base_api_additional_responses,
 )
 
@@ -36,4 +36,4 @@ async def get_upstream(
     response = await get_upstream_response(session=session, upstream_id=body.id)
     if response is None:
         raise ResourceNotExistsError()
-    return CustomORJSONResponse(response)
+    return CustomJSONResponse(response)
diff --git a/src/dstack/_internal/server/routers/templates.py b/src/dstack/_internal/server/routers/templates.py
index 99af9f273c..9349161757 100644
--- a/src/dstack/_internal/server/routers/templates.py
+++ b/src/dstack/_internal/server/routers/templates.py
@@ -6,7 +6,7 @@
 from dstack._internal.server.models import ProjectModel, UserModel
 from dstack._internal.server.security.permissions import ProjectMember
 from dstack._internal.server.services import templates as templates_service
-from dstack._internal.server.utils.routers import CustomORJSONResponse
+from dstack._internal.server.utils.routers import CustomJSONResponse
 
 router = APIRouter(
     prefix="/api/project/{project_name}/templates",
@@ -19,4 +19,4 @@ async def list_templates(
     user_project: Tuple[UserModel, ProjectModel] = Depends(ProjectMember()),
 ):
     _, project = user_project
-    return CustomORJSONResponse(await templates_service.list_templates(project))
+    return CustomJSONResponse(await templates_service.list_templates(project))
diff --git a/src/dstack/_internal/server/routers/users.py b/src/dstack/_internal/server/routers/users.py
index e26ce6dfbe..c5af436b32 100644
--- a/src/dstack/_internal/server/routers/users.py
+++ b/src/dstack/_internal/server/routers/users.py
@@ -18,7 +18,7 @@
 from dstack._internal.server.security.permissions import Authenticated, GlobalAdmin
 from dstack._internal.server.services import events, users
 from dstack._internal.server.utils.routers import (
-    CustomORJSONResponse,
+    CustomJSONResponse,
     get_base_api_additional_responses,
 )
 
@@ -46,7 +46,7 @@ async def list_users(
     if body is None:
         # For backward compatibility
         body = ListUsersRequest()
-    return CustomORJSONResponse(
+    return CustomJSONResponse(
         await users.list_users_for_user(
             session=session,
             user=user,
@@ -68,7 +68,7 @@ async def get_my_user(
     if user.ssh_private_key is None or user.ssh_public_key is None:
         # Generate keys for pre-0.19.33 users
         await users.refresh_ssh_key(session=session, actor=user)
-    return CustomORJSONResponse(users.user_model_to_user_with_creds(user))
+    return CustomJSONResponse(users.user_model_to_user_with_creds(user))
 
 
 @router.post("/get_user", summary="Get user", response_model=UserWithCreds)
@@ -82,7 +82,7 @@ async def get_user(
     )
     if res is None:
         raise ResourceNotExistsError()
-    return CustomORJSONResponse(res)
+    return CustomJSONResponse(res)
 
 
 @router.post("/create", summary="Create user", response_model=User)
@@ -99,7 +99,7 @@ async def create_user(
         active=body.active,
         creator=user,
     )
-    return CustomORJSONResponse(users.user_model_to_user(res))
+    return CustomJSONResponse(users.user_model_to_user(res))
 
 
 @router.post("/update", summary="Update user", response_model=User)
@@ -118,7 +118,7 @@ async def update_user(
     )
     if res is None:
         raise ResourceNotExistsError()
-    return CustomORJSONResponse(users.user_model_to_user(res))
+    return CustomJSONResponse(users.user_model_to_user(res))
 
 
 @router.post("/refresh_ssh_key", summary="Refresh SSH key", response_model=UserWithCreds)
@@ -130,7 +130,7 @@ async def refresh_ssh_key(
     res = await users.refresh_ssh_key(session=session, actor=user, username=body.username)
     if res is None:
         raise ResourceNotExistsError()
-    return CustomORJSONResponse(users.user_model_to_user_with_creds(res))
+    return CustomJSONResponse(users.user_model_to_user_with_creds(res))
 
 
 @router.post("/refresh_token", summary="Refresh token", response_model=UserWithCreds)
@@ -142,7 +142,7 @@ async def refresh_token(
     res = await users.refresh_user_token(session=session, actor=user, username=body.username)
     if res is None:
         raise ResourceNotExistsError()
-    return CustomORJSONResponse(users.user_model_to_user_with_creds(res))
+    return CustomJSONResponse(users.user_model_to_user_with_creds(res))
 
 
 @router.post("/delete", summary="Delete users")
diff --git a/src/dstack/_internal/server/routers/volumes.py b/src/dstack/_internal/server/routers/volumes.py
index 47db390118..8daee52bd4 100644
--- a/src/dstack/_internal/server/routers/volumes.py
+++ b/src/dstack/_internal/server/routers/volumes.py
@@ -17,7 +17,7 @@
 from dstack._internal.server.security.permissions import Authenticated, ProjectMember
 from dstack._internal.server.services.pipelines import PipelineHinterProtocol, get_pipeline_hinter
 from dstack._internal.server.utils.routers import (
-    CustomORJSONResponse,
+    CustomJSONResponse,
     get_base_api_additional_responses,
 )
 
@@ -42,7 +42,7 @@ async def list_volumes(
     The results are paginated. To get the next page, pass `created_at` and `id` of
     the last fleet from the previous page as `prev_created_at` and `prev_id`.
     """
-    return CustomORJSONResponse(
+    return CustomJSONResponse(
         await volumes_services.list_volumes(
             session=session,
             user=user,
@@ -65,7 +65,7 @@ async def list_project_volumes(
     Returns all volumes in the project.
     """
     _, project = user_project
-    return CustomORJSONResponse(
+    return CustomJSONResponse(
         await volumes_services.list_project_volumes(session=session, project=project)
     )
 
@@ -85,7 +85,7 @@ async def get_volume(
     )
     if volume is None:
         raise ResourceNotExistsError()
-    return CustomORJSONResponse(volume)
+    return CustomJSONResponse(volume)
 
 
 @project_router.post("/create", summary="Create volume", response_model=Volume)
@@ -99,7 +99,7 @@ async def create_volume(
     Creates a volume given a volume configuration.
     """
     user, project = user_project
-    return CustomORJSONResponse(
+    return CustomJSONResponse(
         await volumes_services.create_volume(
             session=session,
             project=project,
diff --git a/src/dstack/_internal/server/schemas/events.py b/src/dstack/_internal/server/schemas/events.py
index 3899b1f398..c3af2a5dae 100644
--- a/src/dstack/_internal/server/schemas/events.py
+++ b/src/dstack/_internal/server/schemas/events.py
@@ -3,7 +3,8 @@
 from typing import Annotated, Optional
 from uuid import UUID
 
-from pydantic import Field, root_validator
+from pydantic import Field, model_validator
+from typing_extensions import Self
 
 from dstack._internal.core.models.common import CoreModel
 from dstack._internal.core.models.events import EventTargetType
@@ -21,8 +22,8 @@ class ListEventsRequest(CoreModel):
                 "List of project IDs."
                 " The response will only include events that target the specified projects"
             ),
-            min_items=MIN_FILTER_ITEMS,
-            max_items=MAX_FILTER_ITEMS,
+            min_length=MIN_FILTER_ITEMS,
+            max_length=MAX_FILTER_ITEMS,
         ),
     ] = None
     target_users: Annotated[
@@ -32,8 +33,8 @@ class ListEventsRequest(CoreModel):
                 "List of user IDs."
                 " The response will only include events that target the specified users"
             ),
-            min_items=MIN_FILTER_ITEMS,
-            max_items=MAX_FILTER_ITEMS,
+            min_length=MIN_FILTER_ITEMS,
+            max_length=MAX_FILTER_ITEMS,
         ),
     ] = None
     target_fleets: Annotated[
@@ -43,8 +44,8 @@ class ListEventsRequest(CoreModel):
                 "List of fleet IDs."
                 " The response will only include events that target the specified fleets"
             ),
-            min_items=MIN_FILTER_ITEMS,
-            max_items=MAX_FILTER_ITEMS,
+            min_length=MIN_FILTER_ITEMS,
+            max_length=MAX_FILTER_ITEMS,
         ),
     ] = None
     target_instances: Annotated[
@@ -54,8 +55,8 @@ class ListEventsRequest(CoreModel):
                 "List of instance IDs."
                 " The response will only include events that target the specified instances"
             ),
-            min_items=MIN_FILTER_ITEMS,
-            max_items=MAX_FILTER_ITEMS,
+            min_length=MIN_FILTER_ITEMS,
+            max_length=MAX_FILTER_ITEMS,
         ),
     ] = None
     target_runs: Annotated[
@@ -65,8 +66,8 @@ class ListEventsRequest(CoreModel):
                 "List of run IDs."
                 " The response will only include events that target the specified runs"
             ),
-            min_items=MIN_FILTER_ITEMS,
-            max_items=MAX_FILTER_ITEMS,
+            min_length=MIN_FILTER_ITEMS,
+            max_length=MAX_FILTER_ITEMS,
         ),
     ] = None
     target_jobs: Annotated[
@@ -76,8 +77,8 @@ class ListEventsRequest(CoreModel):
                 "List of job IDs."
                 " The response will only include events that target the specified jobs"
             ),
-            min_items=MIN_FILTER_ITEMS,
-            max_items=MAX_FILTER_ITEMS,
+            min_length=MIN_FILTER_ITEMS,
+            max_length=MAX_FILTER_ITEMS,
         ),
     ] = None
     target_volumes: Annotated[
@@ -87,8 +88,8 @@ class ListEventsRequest(CoreModel):
                 "List of volume IDs."
                 " The response will only include events that target the specified volumes"
             ),
-            min_items=MIN_FILTER_ITEMS,
-            max_items=MAX_FILTER_ITEMS,
+            min_length=MIN_FILTER_ITEMS,
+            max_length=MAX_FILTER_ITEMS,
         ),
     ] = None
     target_gateways: Annotated[
@@ -98,8 +99,8 @@ class ListEventsRequest(CoreModel):
                 "List of gateway IDs."
                 " The response will only include events that target the specified gateways"
             ),
-            min_items=MIN_FILTER_ITEMS,
-            max_items=MAX_FILTER_ITEMS,
+            min_length=MIN_FILTER_ITEMS,
+            max_length=MAX_FILTER_ITEMS,
         ),
     ] = None
     target_secrets: Annotated[
@@ -109,8 +110,8 @@ class ListEventsRequest(CoreModel):
                 "List of secret IDs."
                 " The response will only include events that target the specified secrets"
             ),
-            min_items=MIN_FILTER_ITEMS,
-            max_items=MAX_FILTER_ITEMS,
+            min_length=MIN_FILTER_ITEMS,
+            max_length=MAX_FILTER_ITEMS,
         ),
     ] = None
     within_projects: Annotated[
@@ -121,8 +122,8 @@ class ListEventsRequest(CoreModel):
                 " The response will only include events that target the specified projects"
                 " or any entities within those projects"
             ),
-            min_items=MIN_FILTER_ITEMS,
-            max_items=MAX_FILTER_ITEMS,
+            min_length=MIN_FILTER_ITEMS,
+            max_length=MAX_FILTER_ITEMS,
         ),
     ] = None
     within_fleets: Annotated[
@@ -133,8 +134,8 @@ class ListEventsRequest(CoreModel):
                 " The response will only include events that target the specified fleets"
                 " or instances within those fleets"
             ),
-            min_items=MIN_FILTER_ITEMS,
-            max_items=MAX_FILTER_ITEMS,
+            min_length=MIN_FILTER_ITEMS,
+            max_length=MAX_FILTER_ITEMS,
         ),
     ] = None
     within_runs: Annotated[
@@ -145,8 +146,8 @@ class ListEventsRequest(CoreModel):
                 " The response will only include events that target the specified runs"
                 " or jobs within those runs"
             ),
-            min_items=MIN_FILTER_ITEMS,
-            max_items=MAX_FILTER_ITEMS,
+            min_length=MIN_FILTER_ITEMS,
+            max_length=MAX_FILTER_ITEMS,
         ),
     ] = None
     include_target_types: Annotated[
@@ -157,8 +158,8 @@ class ListEventsRequest(CoreModel):
                 " The response will only include events that have a target"
                 " of one of the specified types"
             ),
-            min_items=MIN_FILTER_ITEMS,
-            max_items=MAX_FILTER_ITEMS,
+            min_length=MIN_FILTER_ITEMS,
+            max_length=MAX_FILTER_ITEMS,
         ),
     ] = None
     actors: Annotated[
@@ -170,8 +171,8 @@ class ListEventsRequest(CoreModel):
                 " performed by the specified users,"
                 " or performed by the system if `null` is specified"
             ),
-            min_items=MIN_FILTER_ITEMS,
-            max_items=MAX_FILTER_ITEMS,
+            min_length=MIN_FILTER_ITEMS,
+            max_length=MAX_FILTER_ITEMS,
         ),
     ] = None
     prev_recorded_at: Optional[datetime] = None
@@ -179,33 +180,33 @@ class ListEventsRequest(CoreModel):
     limit: int = Field(LIST_EVENTS_DEFAULT_LIMIT, ge=1, le=100)
     ascending: bool = False
 
-    @root_validator
-    def _validate_target_filters(cls, values):
+    @model_validator(mode="after")
+    def _validate_target_filters(self) -> Self:
         """
         Raise an error if more than one target_* filter is set. Setting multiple
         target_* filters would always result in an empty response, which might confuse users.
         """
 
-        target_filters = [name for name in cls.__fields__ if name.startswith("target_")]
-        set_filters = [f for f in target_filters if values.get(f) is not None]
+        target_filters = [name for name in type(self).model_fields if name.startswith("target_")]
+        set_filters = [f for f in target_filters if getattr(self, f) is not None]
         if len(set_filters) > 1:
             raise ValueError(
                 f"At most one target_* filter can be set at a time. Got {', '.join(set_filters)}"
             )
-        return values
+        return self
 
-    @root_validator
-    def _validate_within_filters(cls, values):
+    @model_validator(mode="after")
+    def _validate_within_filters(self) -> Self:
         """
         Raise an error if more than one within_* filter is set. Setting multiple
         within_* filters is either redundant or incorrect. Each within_* filter
         may also lead to additional db queries, causing unnecessary load.
         """
 
-        within_filters = [name for name in cls.__fields__ if name.startswith("within_")]
-        set_filters = [f for f in within_filters if values.get(f) is not None]
+        within_filters = [name for name in type(self).model_fields if name.startswith("within_")]
+        set_filters = [f for f in within_filters if getattr(self, f) is not None]
         if len(set_filters) > 1:
             raise ValueError(
                 f"At most one within_* filter can be set at a time. Got {', '.join(set_filters)}"
             )
-        return values
+        return self
diff --git a/src/dstack/_internal/server/schemas/gateways.py b/src/dstack/_internal/server/schemas/gateways.py
index 6ae30a88d1..3c13ba6ae6 100644
--- a/src/dstack/_internal/server/schemas/gateways.py
+++ b/src/dstack/_internal/server/schemas/gateways.py
@@ -1,8 +1,8 @@
-from typing import Annotated, Any, Dict, List, Optional
+from typing import Annotated, List, Optional
 
 from pydantic import Field
 
-from dstack._internal.core.models.common import CoreConfig, CoreModel, generate_dual_core_model
+from dstack._internal.core.models.common import CoreModel
 from dstack._internal.core.models.gateways import (
     ApplyGatewayPlanInput,
     GatewayConfiguration,
@@ -10,13 +10,7 @@
 )
 
 
-class CreateGatewayRequestConfig(CoreConfig):
-    @staticmethod
-    def schema_extra(schema: Dict[str, Any]):
-        pass
-
-
-class CreateGatewayRequest(generate_dual_core_model(CreateGatewayRequestConfig)):
+class CreateGatewayRequest(CoreModel):
     configuration: GatewayConfiguration
 
 
diff --git a/src/dstack/_internal/server/schemas/projects.py b/src/dstack/_internal/server/schemas/projects.py
index c45624f668..f72b20d8b3 100644
--- a/src/dstack/_internal/server/schemas/projects.py
+++ b/src/dstack/_internal/server/schemas/projects.py
@@ -19,7 +19,7 @@ class ListProjectsRequest(CoreModel):
         Optional[str],
         Field(
             description="Include only projects with the name containing `name_pattern`.",
-            regex="^[a-zA-Z0-9-_]*$",
+            pattern="^[a-zA-Z0-9-_]*$",
         ),
     ] = None
     prev_created_at: Annotated[
diff --git a/src/dstack/_internal/server/schemas/runner.py b/src/dstack/_internal/server/schemas/runner.py
index 8363443171..532a8e93fa 100644
--- a/src/dstack/_internal/server/schemas/runner.py
+++ b/src/dstack/_internal/server/schemas/runner.py
@@ -1,9 +1,8 @@
 from base64 import b64decode
 from enum import Enum
-from typing import Dict, List, Optional, Union
+from typing import Any, Dict, List, Optional, Union
 
-from pydantic import Field, validator
-from typing_extensions import Annotated
+from pydantic import field_validator
 
 from dstack._internal.core.models.common import CoreModel, NetworkMode
 from dstack._internal.core.models.repos.remote import RemoteRepoCreds
@@ -33,7 +32,8 @@ class LogEvent(CoreModel):
     """`timestamp` is stored in milliseconds."""
     message: bytes
 
-    @validator("message", pre=True)
+    @field_validator("message", mode="before")
+    @classmethod
     def decode_message(cls, v: Union[str, bytes]) -> bytes:
         if isinstance(v, str):
             return b64decode(v)
@@ -54,73 +54,74 @@ class JobInfoResponse(CoreModel):
     username: str
 
 
+# What the runner is actually sent. This used to be spelled as `Field(include=...)` per field, but
+# pydantic v2 removed `include` from `Field` — it is silently ignored there, which would have sent
+# the runner every field of `Run`, `JobSpec` and `JobSubmission` instead of these subsets. It lives
+# on the model rather than at the call site so that a new caller cannot bypass it.
+#
+# A name the target model does not declare is ignored: `entrypoint` and `gateway` are listed for
+# `job_spec` but `JobSpec` has neither.
+_SUBMIT_BODY_INCLUDE: Dict[str, Any] = {
+    "run": {
+        "id": True,
+        "run_spec": {
+            "run_name",
+            "repo_id",
+            "repo_data",
+            "configuration",
+            "configuration_path",
+        },
+    },
+    "job_spec": {
+        "replica_num",
+        "job_num",
+        "jobs_per_replica",
+        "user",
+        "commands",
+        "entrypoint",
+        "env",
+        "gateway",
+        "single_branch",
+        "max_duration",
+        "ssh_key",
+        "working_dir",
+        "repo_dir",
+        "repo_data",
+        "repo_exists_action",
+        "file_archives",
+    },
+    "job_submission": {"id"},
+    "cluster_info": True,
+    "secrets": True,
+    "repo_credentials": True,
+    "log_quota_hour": True,
+    "run_spec": {
+        "run_name",
+        "repo_id",
+        "repo_data",
+        "configuration",
+        "configuration_path",
+    },
+}
+
+
 class SubmitBody(CoreModel):
-    run: Annotated[
-        Run,
-        Field(
-            include={
-                "id": True,
-                "run_spec": {
-                    "run_name",
-                    "repo_id",
-                    "repo_data",
-                    "configuration",
-                    "configuration_path",
-                },
-            }
-        ),
-    ]
-    job_spec: Annotated[
-        JobSpec,
-        Field(
-            include={
-                "replica_num",
-                "job_num",
-                "jobs_per_replica",
-                "user",
-                "commands",
-                "entrypoint",
-                "env",
-                "gateway",
-                "single_branch",
-                "max_duration",
-                "ssh_key",
-                "working_dir",
-                "repo_dir",
-                "repo_data",
-                "repo_exists_action",
-                "file_archives",
-            }
-        ),
-    ]
-    job_submission: Annotated[
-        JobSubmission,
-        Field(
-            include={
-                "id",
-            }
-        ),
-    ]
-    cluster_info: Annotated[Optional[ClusterInfo], Field(include=True)] = None
-    secrets: Annotated[Optional[Dict[str, str]], Field(include=True)] = None
-    repo_credentials: Annotated[Optional[RemoteRepoCreds], Field(include=True)] = None
-    log_quota_hour: Annotated[Optional[int], Field(include=True)] = None
+    run: Run
+    job_spec: JobSpec
+    job_submission: JobSubmission
+    cluster_info: Optional[ClusterInfo] = None
+    secrets: Optional[Dict[str, str]] = None
+    repo_credentials: Optional[RemoteRepoCreds] = None
+    log_quota_hour: Optional[int] = None
     """Maximum bytes of log output per hour. None means unlimited."""
     # TODO: remove `run_spec` once instances deployed with 0.19.8 or earlier are no longer supported.
-    run_spec: Annotated[
-        RunSpec,
-        Field(
-            include={
-                "run_name",
-                "repo_id",
-                "repo_data",
-                "configuration",
-                "configuration_path",
-            },
-        ),
-    ]
+    run_spec: RunSpec
     """`run_spec` is deprecated in favor of `run.run_spec`."""
 
+    def json_for_runner(self) -> str:
+        """The JSON the runner is sent, restricted to `_SUBMIT_BODY_INCLUDE`."""
+        return self.model_dump_json(include=_SUBMIT_BODY_INCLUDE)
+
 
 class HealthcheckResponse(CoreModel):
     service: str
diff --git a/src/dstack/_internal/server/schemas/runs.py b/src/dstack/_internal/server/schemas/runs.py
index 2163320fd8..09359db304 100644
--- a/src/dstack/_internal/server/schemas/runs.py
+++ b/src/dstack/_internal/server/schemas/runs.py
@@ -43,7 +43,7 @@ class GetRunRequest(CoreModel):
 class GetRunPlanRequest(CoreModel):
     run_spec: RunSpec
     max_offers: Optional[int] = Field(
-        description="The maximum number of offers to return", ge=1, le=10000
+        default=None, description="The maximum number of offers to return", ge=1, le=10000
     )
     full_offers: Annotated[
         bool, Field(description="Return full offers not adjusted by requirements")
diff --git a/src/dstack/_internal/server/schemas/users.py b/src/dstack/_internal/server/schemas/users.py
index b95fcd9cd4..1305185d3c 100644
--- a/src/dstack/_internal/server/schemas/users.py
+++ b/src/dstack/_internal/server/schemas/users.py
@@ -16,7 +16,7 @@ class ListUsersRequest(CoreModel):
         Optional[str],
         Field(
             description="Include only users with the name containing `name_pattern`.",
-            regex="^[a-zA-Z0-9-_]*$",
+            pattern="^[a-zA-Z0-9-_]*$",
         ),
     ] = None
     prev_created_at: Annotated[
diff --git a/src/dstack/_internal/server/services/auth.py b/src/dstack/_internal/server/services/auth.py
index 8ea40994f3..451e36842d 100644
--- a/src/dstack/_internal/server/services/auth.py
+++ b/src/dstack/_internal/server/services/auth.py
@@ -35,7 +35,7 @@ def list_providers() -> list[OAuthProviderInfo]:
 def generate_oauth_state(local_port: Optional[int] = None) -> str:
     value = str(secrets.token_hex(16))
     state = OAuthState(value=value, local_port=local_port)
-    return b64encode(state.json().encode()).decode()
+    return b64encode(state.model_dump_json().encode()).decode()
 
 
 def set_state_cookie(response: Response, state: str):
@@ -71,7 +71,7 @@ def get_next_redirect_url(code: str, state: str) -> Optional[str]:
 
 def _decode_state(state: str) -> Optional[OAuthState]:
     try:
-        return OAuthState.parse_raw(b64decode(state, validate=True).decode())
+        return OAuthState.model_validate_json(b64decode(state, validate=True).decode())
     except Exception as e:
         logger.debug("Exception when decoding OAuth2 state parameter: %s", repr(e))
         return None
diff --git a/src/dstack/_internal/server/services/backends/__init__.py b/src/dstack/_internal/server/services/backends/__init__.py
index 98dcb6b2e8..0c5709f746 100644
--- a/src/dstack/_internal/server/services/backends/__init__.py
+++ b/src/dstack/_internal/server/services/backends/__init__.py
@@ -6,10 +6,9 @@
 from uuid import UUID
 
 from cachetools import TTLCache
-from pydantic import Field, ValidationError
+from pydantic import ValidationError
 from sqlalchemy import delete, update
 from sqlalchemy.ext.asyncio import AsyncSession
-from typing_extensions import Annotated
 
 from dstack._internal.core.backends.base.backend import Backend
 from dstack._internal.core.backends.base.configurator import (
@@ -23,6 +22,7 @@
 from dstack._internal.core.backends.models import (
     AnyBackendConfigWithCreds,
     AnyBackendConfigWithoutCreds,
+    BackendConfigWithCreds,
 )
 from dstack._internal.core.errors import (
     BackendAuthError,
@@ -34,7 +34,6 @@
     ServerClientError,
 )
 from dstack._internal.core.models.backends.base import BackendType
-from dstack._internal.core.models.common import CoreModel
 from dstack._internal.core.models.instances import (
     InstanceOfferWithAvailability,
 )
@@ -48,15 +47,11 @@
 logger = get_logger(__name__)
 
 
-class _BackendConfigWithCreds(CoreModel):
-    __root__: Annotated[AnyBackendConfigWithCreds, Field(..., discriminator="type")]
-
-
 def serialize_source_backend_config(
     config: AnyBackendConfigWithCreds,
 ) -> Tuple[str, Optional[str]]:
     """Split user-intent backend config into non-sensitive and sensitive JSON blobs."""
-    source_config_dict = config.dict()
+    source_config_dict = config.model_dump()
     source_auth = source_config_dict.pop("creds", None)
     source_auth_json = None if source_auth is None else json.dumps(source_auth)
     return json.dumps(source_config_dict), source_auth_json
@@ -218,7 +213,7 @@ def get_source_backend_config_from_backend_model(
             )
             return None
     try:
-        return _BackendConfigWithCreds.parse_obj(source_config_dict).__root__
+        return BackendConfigWithCreds.model_validate(source_config_dict).root
     except ValidationError:
         logger.warning(
             "Failed to validate source config for %s backend. Falling back to stored config.",
diff --git a/src/dstack/_internal/server/services/compute_groups.py b/src/dstack/_internal/server/services/compute_groups.py
index 4d759e0d21..8a05906b29 100644
--- a/src/dstack/_internal/server/services/compute_groups.py
+++ b/src/dstack/_internal/server/services/compute_groups.py
@@ -1,3 +1,4 @@
+from dstack._internal.core.models.common import validate_json_extra_ignore
 from dstack._internal.core.models.compute_groups import ComputeGroup, ComputeGroupProvisioningData
 from dstack._internal.server.models import ComputeGroupModel
 
@@ -17,6 +18,6 @@ def compute_group_model_to_compute_group(compute_group_model: ComputeGroupModel)
 def get_compute_group_provisioning_data(
     compute_group_model: ComputeGroupModel,
 ) -> ComputeGroupProvisioningData:
-    return ComputeGroupProvisioningData.__response__.parse_raw(
-        compute_group_model.provisioning_data
+    return validate_json_extra_ignore(
+        ComputeGroupProvisioningData, compute_group_model.provisioning_data
     )
diff --git a/src/dstack/_internal/server/services/config.py b/src/dstack/_internal/server/services/config.py
index f390200695..81f6776387 100644
--- a/src/dstack/_internal/server/services/config.py
+++ b/src/dstack/_internal/server/services/config.py
@@ -9,6 +9,7 @@
 from dstack._internal.core.backends.models import (
     AnyBackendConfigWithCreds,
     AnyBackendFileConfigWithCreds,
+    BackendConfigWithCreds,
     BackendInfoYAML,
 )
 from dstack._internal.core.errors import (
@@ -221,7 +222,7 @@ def _load_config(self) -> Optional[ServerConfig]:
         except OSError:
             return
         config_dict = yaml.safe_load(content)
-        return ServerConfig.parse_obj(config_dict)
+        return ServerConfig.model_validate(config_dict)
 
     def _save_config(self, config: ServerConfig):
         with open(settings.SERVER_CONFIG_FILE_PATH, "w+") as f:
@@ -261,31 +262,23 @@ async def update_backend_config_yaml(
     await backends_services.update_backend(session=session, project=project, config=config)
 
 
-class _BackendConfigWithCreds(CoreModel):
-    """
-    Model for parsing API and file YAML configs.
-    """
-
-    __root__: Annotated[AnyBackendConfigWithCreds, Field(..., discriminator="type")]
-
-
 def config_yaml_to_backend_config(config_yaml: str) -> AnyBackendConfigWithCreds:
     try:
         config_dict = yaml.safe_load(config_yaml)
     except yaml.YAMLError:
         raise ServerClientError("Error parsing YAML")
     try:
-        backend_config = _BackendConfigWithCreds.parse_obj(config_dict).__root__
+        backend_config = BackendConfigWithCreds.model_validate(config_dict).root
     except ValidationError as e:
         raise ServerClientError(str(e))
     return backend_config
 
 
 def file_config_to_config(file_config: AnyBackendFileConfigWithCreds) -> AnyBackendConfigWithCreds:
-    backend_config_dict = file_config.dict()
-    backend_config = _BackendConfigWithCreds.parse_obj(backend_config_dict)
-    return backend_config.__root__
+    backend_config_dict = file_config.model_dump()
+    backend_config = BackendConfigWithCreds.model_validate(backend_config_dict)
+    return backend_config.root
 
 
 def config_to_yaml(config: CoreModel) -> str:
-    return yaml.dump(config.dict(exclude_none=True), sort_keys=False)
+    return yaml.dump(config.model_dump(exclude_none=True), sort_keys=False)
diff --git a/src/dstack/_internal/server/services/docker.py b/src/dstack/_internal/server/services/docker.py
index 41580e473b..70c3c458a4 100644
--- a/src/dstack/_internal/server/services/docker.py
+++ b/src/dstack/_internal/server/services/docker.py
@@ -5,11 +5,11 @@
 import requests
 from dxf import DXF
 from dxf.exceptions import DXFError
-from pydantic import Field, ValidationError, validator
+from pydantic import Field, ValidationError, field_validator
 from typing_extensions import Annotated
 
 from dstack._internal.core.errors import DockerRegistryError
-from dstack._internal.core.models.common import CoreModel, RegistryAuth
+from dstack._internal.core.models.common import CoreModel, RegistryAuth, validate_json_extra_ignore
 from dstack._internal.server import settings as server_settings
 from dstack._internal.server.utils.common import join_byte_stream_checked
 from dstack._internal.utils.docker import (
@@ -40,7 +40,8 @@ class ImageConfig(CoreModel):
     entrypoint: Annotated[Optional[List[str]], Field(alias="Entrypoint")] = None
     cmd: Annotated[Optional[List[str]], Field(alias="Cmd")] = None
 
-    @validator("user")
+    @field_validator("user")
+    @classmethod
     def normalize_user(cls, v: Optional[str]) -> Optional[str]:
         # If USER is not set, the corresponding field may be missing or set to an empty string
         if v == "":
@@ -51,7 +52,8 @@ def normalize_user(cls, v: Optional[str]) -> Optional[str]:
 class ImageConfigObject(CoreModel):
     config: ImageConfig = ImageConfig()
 
-    @validator("config", pre=True)
+    @field_validator("config", mode="before")
+    @classmethod
     def config_set_default_if_null(cls, value):
         return ImageConfig() if value is None else value
 
@@ -83,14 +85,17 @@ def get_image_config(image_name: str, registry_auth: Optional[RegistryAuth]) ->
             manifest_resp = registry_client.get_manifest(
                 alias=image.digest or image.tag, platform=DEFAULT_PLATFORM
             )
-            manifest = ImageManifest.__response__.parse_raw(manifest_resp)
+            assert isinstance(manifest_resp, str), (
+                "get_manifest() returns the manifest JSON when `platform` is given"
+            )
+            manifest = validate_json_extra_ignore(ImageManifest, manifest_resp)
             config_stream = registry_client.pull_blob(manifest.config.digest)
             config_resp = join_byte_stream_checked(config_stream, MAX_CONFIG_OBJECT_SIZE)  # type: ignore[arg-type]
             if config_resp is None:
                 raise DockerRegistryError(
                     f"Image config object exceeds the size limit of {MAX_CONFIG_OBJECT_SIZE} bytes"
                 )
-            return ImageConfigObject.__response__.parse_raw(config_resp)
+            return validate_json_extra_ignore(ImageConfigObject, config_resp)
 
         except (DXFError, requests.RequestException, ValidationError) as e:
             raise DockerRegistryError(e)
diff --git a/src/dstack/_internal/server/services/encryption/keys/aes.py b/src/dstack/_internal/server/services/encryption/keys/aes.py
index 4c6e08064e..56a533bc91 100644
--- a/src/dstack/_internal/server/services/encryption/keys/aes.py
+++ b/src/dstack/_internal/server/services/encryption/keys/aes.py
@@ -3,7 +3,7 @@
 from typing import Literal
 
 from cryptography.hazmat.primitives.ciphers.aead import AESGCM
-from pydantic import Field, validator
+from pydantic import Field, field_validator
 from typing_extensions import Annotated
 
 from dstack._internal.core.models.common import CoreModel
@@ -15,13 +15,15 @@ class AESEncryptionKeyConfig(CoreModel):
     name: Annotated[str, Field(description="The key name for key identification")]
     secret: Annotated[str, Field(description="Base64-encoded AES-256 key")]
 
-    @validator("name")
+    @field_validator("name")
+    @classmethod
     def validate_name(cls, v):
         if not v.isalnum():
             raise ValueError("Key name must be alphanumeric")
         return v
 
-    @validator("secret")
+    @field_validator("secret")
+    @classmethod
     def validate_secret(cls, v):
         try:
             key = b64decode(v, validate=True)
diff --git a/src/dstack/_internal/server/services/fleets.py b/src/dstack/_internal/server/services/fleets.py
index 4e93e7c2e1..2c6a924c50 100644
--- a/src/dstack/_internal/server/services/fleets.py
+++ b/src/dstack/_internal/server/services/fleets.py
@@ -16,7 +16,7 @@
     ResourceExistsError,
     ServerClientError,
 )
-from dstack._internal.core.models.common import ApplyAction, CoreModel
+from dstack._internal.core.models.common import ApplyAction, CoreModel, validate_json_extra_ignore
 from dstack._internal.core.models.envs import Env
 from dstack._internal.core.models.fleets import (
     ApplyFleetPlanInput,
@@ -415,7 +415,6 @@ async def get_plan(
     user: UserModel,
     spec: FleetSpec,
 ) -> FleetPlan:
-    # Spec must be copied by parsing to calculate merged_profile
     effective_spec = copy_model(spec)
     effective_spec = await apply_plugin_policies(
         user=user.name,
@@ -907,7 +906,7 @@ def fleet_model_to_fleet(
 
 
 def get_fleet_spec(fleet_model: FleetModel) -> FleetSpec:
-    return FleetSpec.__response__.parse_raw(fleet_model.spec)
+    return validate_json_extra_ignore(FleetSpec, fleet_model.spec)
 
 
 async def generate_fleet_name(session: AsyncSession, project: ProjectModel) -> str:
@@ -982,8 +981,8 @@ def get_fleet_master_instance_provisioning_data(
                 and not instance_model.deleted
                 and instance_model.job_provisioning_data is not None
             ):
-                return JobProvisioningData.__response__.parse_raw(
-                    instance_model.job_provisioning_data
+                return validate_json_extra_ignore(
+                    JobProvisioningData, instance_model.job_provisioning_data
                 )
 
     return None
@@ -1043,7 +1042,7 @@ async def _create_fleet(
             name=spec.configuration.name,
             project=project,
             status=FleetStatus.ACTIVE,
-            spec=spec.json(),
+            spec=spec.model_dump_json(),
             instances=[],
             created_at=now,
             last_processed_at=now,
@@ -1134,7 +1133,7 @@ async def _update_fleet(
 
     _check_can_update_fleet_spec(fleet_sensitive.spec, spec)
 
-    fleet_model.spec = spec.json()
+    fleet_model.spec = spec.model_dump_json()
     # Reset consolidation attempt so the next pipeline pass picks up the spec change promptly.
     fleet_model.consolidation_attempt = 0
 
@@ -1363,6 +1362,11 @@ def _remove_fleet_spec_sensitive_info(spec: FleetSpec):
 
 
 def _validate_fleet_spec_and_set_defaults(spec: FleetSpec):
+    # Callers do not reparse afterwards, so the defaults set here must not touch any field that
+    # `ProfileParams` also declares — `spec.merged_profile` is computed at parse time and would
+    # silently keep the pre-default value. Only `configuration.resources` is written, which
+    # `ProfileParams` does not declare.
+    # TODO: Make callers reparse if this changes.
     if spec.configuration.name is not None:
         validate_dstack_resource_name(spec.configuration.name)
     _validate_fleet_configuration_subtype_specific_fields(spec.configuration)
@@ -1391,9 +1395,14 @@ def _validate_fleet_configuration_subtype_specific_fields(conf: FleetConfigurati
         subtype = "Backend"
         props_model = SSHFleetConfigurationProps
     non_default_fields: list[str] = []
-    for field in props_model.__fields__.values():
-        if getattr(conf, field.name) != field.default:
-            non_default_fields.append(field.name)
+    for name, field in props_model.model_fields.items():
+        # `FieldInfo` has no `.name` in pydantic v2, and `.default` is `PydanticUndefined` rather
+        # than `None` for a required field — comparing against it directly would silently report
+        # every required field as non-default. No props field is required today, but that would
+        # arm itself the moment one is added.
+        default = None if field.is_required() else field.get_default(call_default_factory=True)
+        if getattr(conf, name) != default:
+            non_default_fields.append(name)
     if non_default_fields:
         raise ServerClientError(
             f"{subtype} fleet configuration does not support the following fields:"
diff --git a/src/dstack/_internal/server/services/gateways/__init__.py b/src/dstack/_internal/server/services/gateways/__init__.py
index 4cf2f30734..a59fdf3451 100644
--- a/src/dstack/_internal/server/services/gateways/__init__.py
+++ b/src/dstack/_internal/server/services/gateways/__init__.py
@@ -29,7 +29,11 @@
     SSHError,
 )
 from dstack._internal.core.models.backends.base import BackendType
-from dstack._internal.core.models.common import ApplyAction, EntityReference
+from dstack._internal.core.models.common import (
+    ApplyAction,
+    EntityReference,
+    validate_json_extra_ignore,
+)
 from dstack._internal.core.models.gateways import (
     GATEWAY_REPLICAS_DEFAULT,
     AnyGatewayRouterConfig,
@@ -216,7 +220,7 @@ def create_gateway_compute_model(
         gateway_id=gateway_id,
         backend_id=backend_id,
         replica_num=replica_num,
-        configuration=compute_configuration.json(),
+        configuration=compute_configuration.model_dump_json(),
         ssh_private_key=gateway_ssh_private_key,
         ssh_public_key=gateway_ssh_public_key,
         status=GatewayReplicaStatus.SUBMITTED,
@@ -270,7 +274,7 @@ async def create_gateway(
             project_id=project.id,
             backend_id=backend_model.id,
             wildcard_domain=configuration.domain,
-            configuration=configuration.json(),
+            configuration=configuration.model_dump_json(),
             status=GatewayStatus.SUBMITTED,
             desired_replica_count=(
                 configuration.replicas
@@ -416,7 +420,7 @@ async def set_gateway_wildcard_domain(
             if gateway.configuration is not None:
                 conf = get_gateway_configuration(gateway)
                 conf.domain = wildcard_domain
-                gateway.configuration = conf.json()
+                gateway.configuration = conf.model_dump_json()
             events.emit(
                 session,
                 f"Gateway wildcard domain changed {old_domain!r} -> {gateway.wildcard_domain!r}",
@@ -808,8 +812,8 @@ def _get_gateway_compute_router_config(
 ) -> Optional[AnyGatewayRouterConfig]:
     if compute.configuration is None:  # pre-0.18.2 gateway
         return None  # gateway routers introduced in 0.19.38
-    compute_config: GatewayComputeConfiguration = (
-        GatewayComputeConfiguration.__response__.parse_raw(compute.configuration)
+    compute_config: GatewayComputeConfiguration = validate_json_extra_ignore(
+        GatewayComputeConfiguration, compute.configuration
     )
     return compute_config.router
 
@@ -855,7 +859,7 @@ def get_gateway_compute_models(gateway_model: GatewayModel) -> List[GatewayCompu
 
 def get_gateway_configuration(gateway_model: GatewayModel) -> GatewayConfiguration:
     if gateway_model.configuration is not None:
-        return GatewayConfiguration.__response__.parse_raw(gateway_model.configuration)
+        return validate_json_extra_ignore(GatewayConfiguration, gateway_model.configuration)
     # Handle gateways created before GatewayConfiguration was introduced
     return GatewayConfiguration(
         name=gateway_model.name,
@@ -871,7 +875,9 @@ def get_gateway_compute_configuration(
     gateway_model: GatewayModel,
 ) -> GatewayComputeConfiguration:
     if gateway_compute.configuration is not None:
-        return GatewayComputeConfiguration.__response__.parse_raw(gateway_compute.configuration)
+        return validate_json_extra_ignore(
+            GatewayComputeConfiguration, gateway_compute.configuration
+        )
     # Handle gateways created before GatewayComputeConfiguration was introduced
     gateway_configuration = get_gateway_configuration(gateway_model)
     return GatewayComputeConfiguration(
@@ -1079,7 +1085,7 @@ async def apply_plan(
                 if new_configuration.replicas is not None
                 else GATEWAY_REPLICAS_DEFAULT
             )
-        gateway_model.configuration = new_configuration.json()
+        gateway_model.configuration = new_configuration.model_dump_json()
         gateway_model.last_update_at = get_current_datetime()
         events.emit(
             session,
diff --git a/src/dstack/_internal/server/services/gateways/client.py b/src/dstack/_internal/server/services/gateways/client.py
index 4abd98811c..7dabe87536 100644
--- a/src/dstack/_internal/server/services/gateways/client.py
+++ b/src/dstack/_internal/server/services/gateways/client.py
@@ -3,7 +3,7 @@
 from typing import Optional
 
 import httpx
-from pydantic import parse_obj_as
+from pydantic import TypeAdapter
 
 from dstack._internal.core.consts import DSTACK_RUNNER_SSH_PORT
 from dstack._internal.core.errors import GatewayError
@@ -60,10 +60,10 @@ async def register_service(
             "auth": auth,
             "client_max_body_size": client_max_body_size,
             "options": options,
-            "rate_limits": [limit.dict() for limit in rate_limits],
+            "rate_limits": [limit.model_dump() for limit in rate_limits],
             "ssh_private_key": ssh_private_key,
             "has_router_replica": has_router_replica,
-            "router": router.dict() if router is not None else None,
+            "router": router.model_dump() if router is not None else None,
         }
         resp = await self._client.post(
             self._url(f"/api/registry/{project}/services/register"), json=payload
@@ -95,7 +95,7 @@ async def register_replica(
         payload = {
             "job_id": job_submission.id.hex,
             "app_port": get_service_port(job_spec, run.run_spec.configuration),
-            "ssh_head_proxy": ssh_head_proxy.dict() if ssh_head_proxy is not None else None,
+            "ssh_head_proxy": ssh_head_proxy.model_dump() if ssh_head_proxy is not None else None,
             "ssh_head_proxy_private_key": ssh_head_proxy_private_key,
         }
         jpd = job_submission.job_provisioning_data
@@ -108,7 +108,7 @@ async def register_replica(
                 {
                     "ssh_port": jpd.ssh_port,
                     "ssh_host": f"{jpd.username}@{jpd.hostname}",
-                    "ssh_proxy": jpd.ssh_proxy.dict() if jpd.ssh_proxy is not None else None,
+                    "ssh_proxy": jpd.ssh_proxy.model_dump() if jpd.ssh_proxy is not None else None,
                 }
             )
         else:
@@ -124,7 +124,7 @@ async def register_replica(
                         hostname=jpd.hostname,
                         username=jpd.username,
                         port=jpd.ssh_port,
-                    ).dict(),
+                    ).model_dump(),
                     "ssh_proxy_private_key": instance_project_ssh_private_key,
                 }
             )
@@ -196,7 +196,7 @@ async def collect_stats(self) -> list[ServiceStats]:
             # Avoid errors if gateway is updated to new format and current server replica isn't.
             # TODO: remove after a few releases
             return []
-        return parse_obj_as(list[ServiceStats], resp_data)
+        return TypeAdapter(list[ServiceStats]).validate_python(resp_data)
 
     def _url(self, path: str) -> str:
         return f"{self.base_url}/{path.lstrip('/')}"
diff --git a/src/dstack/_internal/server/services/instances.py b/src/dstack/_internal/server/services/instances.py
index 71cd3aac2d..129aaf4c07 100644
--- a/src/dstack/_internal/server/services/instances.py
+++ b/src/dstack/_internal/server/services/instances.py
@@ -16,7 +16,7 @@
 from dstack._internal.core.backends.features import BACKENDS_WITH_MULTINODE_SUPPORT
 from dstack._internal.core.errors import ResourceNotExistsError
 from dstack._internal.core.models.backends.base import BackendType
-from dstack._internal.core.models.common import EntityReference
+from dstack._internal.core.models.common import EntityReference, validate_json_extra_ignore
 from dstack._internal.core.models.envs import Env
 from dstack._internal.core.models.health import HealthCheck, HealthEvent, HealthStatus
 from dstack._internal.core.models.instances import (
@@ -63,6 +63,7 @@
 from dstack._internal.server.services.projects import list_user_project_models
 from dstack._internal.server.services.runner.client import ShimClient
 from dstack._internal.utils import common as common_utils
+from dstack._internal.utils.common import get_or_error
 from dstack._internal.utils.logging import get_logger
 
 logger = get_logger(__name__)
@@ -303,31 +304,35 @@ def dcgm_health_response_to_health_check(
 def get_instance_health_response(
     instance_health_check_model: InstanceHealthCheckModel,
 ) -> InstanceHealthResponse:
-    return InstanceHealthResponse.__response__.parse_raw(instance_health_check_model.response)
+    return validate_json_extra_ignore(InstanceHealthResponse, instance_health_check_model.response)
 
 
 def get_instance_provisioning_data(instance_model: InstanceModel) -> Optional[JobProvisioningData]:
     if instance_model.job_provisioning_data is None:
         return None
-    return JobProvisioningData.__response__.parse_raw(instance_model.job_provisioning_data)
+    return validate_json_extra_ignore(JobProvisioningData, instance_model.job_provisioning_data)
 
 
 def get_instance_offer(instance_model: InstanceModel) -> Optional[InstanceOfferWithAvailability]:
     if instance_model.offer is None:
         return None
-    return InstanceOfferWithAvailability.__response__.parse_raw(instance_model.offer)
+    return validate_json_extra_ignore(
+        InstanceOfferWithAvailability, get_or_error(instance_model.offer)
+    )
 
 
 def get_instance_configuration(instance_model: InstanceModel) -> InstanceConfiguration:
-    return InstanceConfiguration.__response__.parse_raw(instance_model.instance_configuration)
+    return validate_json_extra_ignore(
+        InstanceConfiguration, get_or_error(instance_model.instance_configuration)
+    )
 
 
 def get_instance_profile(instance_model: InstanceModel) -> Profile:
-    return Profile.__response__.parse_raw(instance_model.profile)
+    return validate_json_extra_ignore(Profile, get_or_error(instance_model.profile))
 
 
 def get_instance_requirements(instance_model: InstanceModel) -> Requirements:
-    return Requirements.__response__.parse_raw(instance_model.requirements)
+    return validate_json_extra_ignore(Requirements, get_or_error(instance_model.requirements))
 
 
 def is_ssh_instance(instance_model: InstanceModel) -> bool:
@@ -357,7 +362,7 @@ def get_instance_remote_connection_info(
 ) -> Optional[RemoteConnectionInfo]:
     if instance_model.remote_connection_info is None:
         return None
-    return RemoteConnectionInfo.__response__.parse_raw(instance_model.remote_connection_info)
+    return validate_json_extra_ignore(RemoteConnectionInfo, instance_model.remote_connection_info)
 
 
 def get_instance_ssh_private_keys(instance_model: InstanceModel) -> tuple[str, Optional[str]]:
@@ -583,7 +588,7 @@ def instance_matches_constraints(
     if requirements is not None:
         if instance.offer is None:
             return False
-        offer = InstanceOffer.__response__.parse_raw(instance.offer)
+        offer = validate_json_extra_ignore(InstanceOffer, instance.offer)
         catalog_item = offer_to_catalog_item(offer)
         if not gpuhunt.matches(catalog_item, q=requirements_to_query_filter(requirements)):
             return False
@@ -912,9 +917,9 @@ def create_instance_model(
         last_processed_at=now,
         status=InstanceStatus.PENDING,
         unreachable=False,
-        profile=profile.json(),
-        requirements=requirements.json(),
-        instance_configuration=instance_config.json(),
+        profile=profile.model_dump_json(),
+        requirements=requirements.model_dump_json(),
+        instance_configuration=instance_config.model_dump_json(),
         termination_policy=termination_policy,
         termination_idle_time=termination_idle_time,
         total_blocks=None if blocks == "auto" else blocks,
@@ -987,9 +992,9 @@ async def create_ssh_instance_model(
         started_at=common_utils.get_current_datetime(),
         status=InstanceStatus.PENDING,
         unreachable=False,
-        job_provisioning_data=remote.json(),
-        remote_connection_info=remote_connection_info.json(),
-        offer=offer.json(),
+        job_provisioning_data=remote.model_dump_json(),
+        remote_connection_info=remote_connection_info.model_dump_json(),
+        offer=offer.model_dump_json(),
         region=offer.region,
         price=offer.price,
         termination_policy=TerminationPolicy.DONT_DESTROY,
diff --git a/src/dstack/_internal/server/services/jobs/__init__.py b/src/dstack/_internal/server/services/jobs/__init__.py
index e0c99221b3..24ceec355d 100644
--- a/src/dstack/_internal/server/services/jobs/__init__.py
+++ b/src/dstack/_internal/server/services/jobs/__init__.py
@@ -17,6 +17,7 @@
     SSHError,
 )
 from dstack._internal.core.models.backends.base import BackendType
+from dstack._internal.core.models.common import validate_json_extra_ignore
 from dstack._internal.core.models.configurations import RunConfigurationType
 from dstack._internal.core.models.runs import (
     ImagePullProgress,
@@ -282,23 +283,23 @@ def job_model_to_job_submission(
 def get_job_provisioning_data(job_model: JobModel) -> Optional[JobProvisioningData]:
     if job_model.job_provisioning_data is None:
         return None
-    return JobProvisioningData.__response__.parse_raw(job_model.job_provisioning_data)
+    return validate_json_extra_ignore(JobProvisioningData, job_model.job_provisioning_data)
 
 
 def get_job_runtime_data(job_model: JobModel) -> Optional[JobRuntimeData]:
     if job_model.job_runtime_data is None:
         return None
-    return JobRuntimeData.__response__.parse_raw(job_model.job_runtime_data)
+    return validate_json_extra_ignore(JobRuntimeData, job_model.job_runtime_data)
 
 
 def _get_image_pull_progress(job_model: JobModel) -> Optional[ImagePullProgress]:
     if job_model.image_pull_progress is None:
         return None
-    return ImagePullProgress.__response__.parse_raw(job_model.image_pull_progress)
+    return validate_json_extra_ignore(ImagePullProgress, job_model.image_pull_progress)
 
 
 def get_job_spec(job_model: JobModel) -> JobSpec:
-    return JobSpec.__response__.parse_raw(job_model.job_spec_data)
+    return validate_json_extra_ignore(JobSpec, job_model.job_spec_data)
 
 
 def delay_job_instance_termination(job_model: JobModel):
diff --git a/src/dstack/_internal/server/services/jobs/configurators/base.py b/src/dstack/_internal/server/services/jobs/configurators/base.py
index c4959a41c7..5ec38790dd 100644
--- a/src/dstack/_internal/server/services/jobs/configurators/base.py
+++ b/src/dstack/_internal/server/services/jobs/configurators/base.py
@@ -1,3 +1,4 @@
+import json
 import shlex
 import sys
 import threading
@@ -5,7 +6,6 @@
 from pathlib import PurePosixPath
 from typing import Dict, List, Optional
 
-import orjson
 from cachetools import TTLCache, cached
 
 from dstack._internal import settings
@@ -446,7 +446,7 @@ def interpolate_job_volumes(
     job_volumes = []
     for mount_point in run_volumes:
         if not isinstance(mount_point, VolumeMountPoint):
-            job_volumes.append(mount_point.copy())
+            job_volumes.append(mount_point.model_copy())
             continue
         if isinstance(mount_point.name, str):
             names = [mount_point.name]
@@ -480,13 +480,13 @@ def _probe_config_to_spec(c: ProbeConfig) -> ProbeSpec:
 
 
 def _openai_model_probe_spec(model_name: str, prefix: str) -> ProbeSpec:
-    body = orjson.dumps(
+    body = json.dumps(
         {
             "model": model_name,
             "messages": [{"role": "user", "content": "hi"}],
             "max_tokens": 1,
         }
-    ).decode("utf-8")
+    )
     return ProbeSpec(
         type="http",
         method="post",
diff --git a/src/dstack/_internal/server/services/logs/filelog.py b/src/dstack/_internal/server/services/logs/filelog.py
index e4289805c6..9f145f1d22 100644
--- a/src/dstack/_internal/server/services/logs/filelog.py
+++ b/src/dstack/_internal/server/services/logs/filelog.py
@@ -4,6 +4,7 @@
 from uuid import UUID
 
 from dstack._internal.core.errors import ServerClientError
+from dstack._internal.core.models.common import validate_json_extra_ignore
 from dstack._internal.core.models.logs import (
     JobSubmissionLogs,
     LogEvent,
@@ -72,7 +73,7 @@ def _poll_logs_ascending(
                     current_line += 1
 
                     try:
-                        log_event = LogEvent.__response__.parse_raw(line)
+                        log_event = validate_json_extra_ignore(LogEvent, line)
                     except Exception:
                         # Skip malformed lines
                         continue
@@ -109,7 +110,7 @@ def _poll_logs_descending(
             for line_bytes, line_start_offset in line_generator:
                 try:
                     line_str = line_bytes.decode("utf-8")
-                    log_event = LogEvent.__response__.parse_raw(line_str)
+                    log_event = validate_json_extra_ignore(LogEvent, line_str)
                 except Exception:
                     continue  # Skip malformed lines
 
@@ -221,7 +222,7 @@ def _write_logs(self, log_file_path: Path, log_events: List[RunnerLogEvent]) ->
         log_events_parsed = [self._runner_log_event_to_log_event(event) for event in log_events]
         log_file_path.parent.mkdir(exist_ok=True, parents=True)
         with open(log_file_path, "a") as f:
-            f.writelines(log.json() + "\n" for log in log_events_parsed)
+            f.writelines(log.model_dump_json() + "\n" for log in log_events_parsed)
 
     def _get_log_file_path(
         self,
diff --git a/src/dstack/_internal/server/services/offers.py b/src/dstack/_internal/server/services/offers.py
index a35f1bef93..0aaa7f3a17 100644
--- a/src/dstack/_internal/server/services/offers.py
+++ b/src/dstack/_internal/server/services/offers.py
@@ -185,7 +185,7 @@ def get_instance_offer_with_restricted_az(
     instance_offer: InstanceOfferWithAvailability,
     master_job_provisioning_data: Optional[JobProvisioningData],
 ) -> InstanceOfferWithAvailability:
-    instance_offer = instance_offer.copy()
+    instance_offer = instance_offer.model_copy()
     if (
         master_job_provisioning_data is not None
         and master_job_provisioning_data.availability_zone is not None
@@ -233,7 +233,7 @@ def _filter_offers(
         if availability_zones is not None:
             if offer.availability_zones is None:
                 continue
-            new_offer = offer.copy()
+            new_offer = offer.model_copy()
             new_offer.availability_zones = [
                 z for z in offer.availability_zones if z in availability_zones
             ]
@@ -267,6 +267,6 @@ def _get_shareable_offers(
         divisible, total_blocks = is_divisible_into_blocks(cpu_count, gpu_count, blocks)
         if not divisible:
             continue
-        new_offer = offer.copy()
+        new_offer = offer.model_copy()
         new_offer.total_blocks = total_blocks
         yield (backend, new_offer)
diff --git a/src/dstack/_internal/server/services/placement.py b/src/dstack/_internal/server/services/placement.py
index 3292b70293..6544442fff 100644
--- a/src/dstack/_internal/server/services/placement.py
+++ b/src/dstack/_internal/server/services/placement.py
@@ -11,6 +11,7 @@
     generate_unique_placement_group_name,
 )
 from dstack._internal.core.errors import BackendError, PlacementGroupNotSupportedError
+from dstack._internal.core.models.common import validate_json_extra_ignore
 from dstack._internal.core.models.instances import InstanceOffer
 from dstack._internal.core.models.placement import (
     PlacementGroup,
@@ -50,7 +51,9 @@ def placement_group_model_to_placement_group_optional(
 def get_placement_group_configuration(
     placement_group_model: PlacementGroupModel,
 ) -> PlacementGroupConfiguration:
-    return PlacementGroupConfiguration.__response__.parse_raw(placement_group_model.configuration)
+    return validate_json_extra_ignore(
+        PlacementGroupConfiguration, placement_group_model.configuration
+    )
 
 
 def get_placement_group_provisioning_data(
@@ -58,8 +61,8 @@ def get_placement_group_provisioning_data(
 ) -> Optional[PlacementGroupProvisioningData]:
     if placement_group_model.provisioning_data is None:
         return None
-    return PlacementGroupProvisioningData.__response__.parse_raw(
-        placement_group_model.provisioning_data
+    return validate_json_extra_ignore(
+        PlacementGroupProvisioningData, placement_group_model.provisioning_data
     )
 
 
@@ -170,7 +173,7 @@ async def create_placement_group(
             backend=master_instance_offer.backend,
             region=master_instance_offer.region,
             placement_strategy=PlacementStrategy.CLUSTER,
-        ).json(),
+        ).model_dump_json(),
     )
     placement_group = placement_group_model_to_placement_group(placement_group_model)
     logger.debug(
@@ -214,5 +217,5 @@ async def create_placement_group(
         placement_group.configuration.backend.value,
         placement_group.configuration.region,
     )
-    placement_group_model.provisioning_data = pgpd.json()
+    placement_group_model.provisioning_data = pgpd.model_dump_json()
     return placement_group_model
diff --git a/src/dstack/_internal/server/services/proxy/repo.py b/src/dstack/_internal/server/services/proxy/repo.py
index e2a8d3c117..17c006b916 100644
--- a/src/dstack/_internal/server/services/proxy/repo.py
+++ b/src/dstack/_internal/server/services/proxy/repo.py
@@ -7,6 +7,7 @@
 
 import dstack._internal.server.services.jobs as jobs_services
 from dstack._internal.core.consts import DSTACK_RUNNER_SSH_PORT
+from dstack._internal.core.models.common import validate_json_extra_ignore
 from dstack._internal.core.models.configurations import ServiceConfiguration
 from dstack._internal.core.models.instances import SSHConnectionParams
 from dstack._internal.core.models.runs import (
@@ -34,6 +35,8 @@
 from dstack._internal.server.settings import DEFAULT_SERVICE_CLIENT_MAX_BODY_SIZE
 from dstack._internal.utils.common import get_or_error
 
+_ANY_MODEL_ADAPTER = pydantic.TypeAdapter(AnyModel)
+
 
 class ServerProxyRepo(BaseProxyRepo):
     """
@@ -78,8 +81,8 @@ async def get_service(self, project_name: str, run_name: str) -> Optional[Servic
         router = run_spec.configuration.router
         replicas = []
         for job in jobs:
-            jpd: JobProvisioningData = JobProvisioningData.__response__.parse_raw(
-                job.job_provisioning_data
+            jpd: JobProvisioningData = validate_json_extra_ignore(
+                JobProvisioningData, get_or_error(job.job_provisioning_data)
             )
             assert jpd.hostname is not None
             assert jpd.ssh_port is not None
@@ -153,12 +156,14 @@ async def list_models(self, project_name: str) -> List[ChatModel]:
         )
         models = []
         for run in res.scalars().all():
-            service_spec: ServiceSpec = ServiceSpec.__response__.parse_raw(run.service_spec)
+            service_spec: ServiceSpec = validate_json_extra_ignore(
+                ServiceSpec, get_or_error(run.service_spec)
+            )
             model_spec = service_spec.model
             model_options_obj = service_spec.options.get("openai", {}).get("model")
             if model_spec is None or model_options_obj is None:
                 continue
-            model_options = pydantic.parse_obj_as(AnyModel, model_options_obj)  # type: ignore[arg-type]
+            model_options = _ANY_MODEL_ADAPTER.validate_python(model_options_obj)
             model = ChatModel(
                 project_name=project_name,
                 name=model_spec.name,
diff --git a/src/dstack/_internal/server/services/repos.py b/src/dstack/_internal/server/services/repos.py
index fd5bf77f38..a83c35064a 100644
--- a/src/dstack/_internal/server/services/repos.py
+++ b/src/dstack/_internal/server/services/repos.py
@@ -12,6 +12,7 @@
     ResourceNotExistsError,
     ServerClientError,
 )
+from dstack._internal.core.models.common import validate_extra_ignore
 from dstack._internal.core.models.repos import (
     AnyRepoInfo,
     RepoHead,
@@ -58,7 +59,7 @@ async def get_repo(
     if repo is None:
         return None
     if not include_creds or repo.type != RepoType.REMOTE:
-        return RepoHeadWithCreds.parse_obj(repo_model_to_repo_head(repo))
+        return RepoHeadWithCreds.model_validate(repo_model_to_repo_head(repo).model_dump())
     repo_creds = await get_repo_creds(
         session=session,
         repo=repo,
@@ -130,7 +131,7 @@ async def create_repo(
         project_id=project.id,
         name=repo_id,
         type=RepoType(repo_info.repo_type),
-        info=repo_info.json(),
+        info=repo_info.model_dump_json(),
     )
     try:
         async with session.begin_nested():
@@ -154,7 +155,7 @@ async def update_repo(
             RepoModel.name == repo_id,
         )
         .values(
-            info=repo_info.json(),
+            info=repo_info.model_dump_json(),
         )
     )
     await session.commit()
@@ -221,7 +222,7 @@ async def create_repo_creds(
     repo_creds = RepoCredsModel(
         repo_id=repo.id,
         user_id=user.id,
-        creds=DecryptedString(plaintext=creds.json()),
+        creds=DecryptedString(plaintext=creds.model_dump_json()),
     )
     try:
         async with session.begin_nested():
@@ -245,7 +246,7 @@ async def update_repo_creds(
             RepoCredsModel.user_id == user.id,
         )
         .values(
-            creds=DecryptedString(plaintext=creds.json()),
+            creds=DecryptedString(plaintext=creds.model_dump_json()),
         )
     )
     await session.commit()
@@ -343,11 +344,12 @@ async def get_code_model(
 
 
 def repo_model_to_repo_head(repo_model: RepoModel) -> RepoHead:
-    return RepoHead.__response__.parse_obj(
+    return validate_extra_ignore(
+        RepoHead,
         {
             "repo_id": repo_model.name,
             "repo_info": json.loads(repo_model.info),
-        }
+        },
     )
 
 
@@ -359,10 +361,11 @@ def repo_model_to_repo_head_with_creds(
         repo_creds_raw = repo_model.creds
     else:
         repo_creds_raw = repo_creds_model.creds.plaintext
-    return RepoHeadWithCreds.__response__.parse_obj(
+    return validate_extra_ignore(
+        RepoHeadWithCreds,
         {
             "repo_id": repo_model.name,
             "repo_info": json.loads(repo_model.info),
             "repo_creds": json.loads(repo_creds_raw) if repo_creds_raw else None,
-        }
+        },
     )
diff --git a/src/dstack/_internal/server/services/requirements/combine.py b/src/dstack/_internal/server/services/requirements/combine.py
index 53cec3457e..4ad168586e 100644
--- a/src/dstack/_internal/server/services/requirements/combine.py
+++ b/src/dstack/_internal/server/services/requirements/combine.py
@@ -86,7 +86,7 @@ def _combine_backend_options(
         if opt.type in by_type:
             by_type[opt.type] = by_type[opt.type].combine(opt)
         else:
-            by_type[opt.type] = opt.copy(deep=True)
+            by_type[opt.type] = opt.model_copy(deep=True)
     return list(by_type.values())
 
 
@@ -251,10 +251,10 @@ def _combine_models_optional(
 ) -> Optional[_ModelT]:
     if value1 is None:
         if value2 is not None:
-            return value2.copy(deep=True)
+            return value2.model_copy(deep=True)
         return None
     if value2 is None:
-        return value1.copy(deep=True)
+        return value1.model_copy(deep=True)
     return combiner(value1, value2)
 
 
@@ -279,8 +279,8 @@ def _combine_copy_model_list_optional(
 ) -> Optional[List[_ModelT]]:
     if value1 is None:
         if value2 is not None:
-            return [item.copy(deep=True) for item in value2]
+            return [item.model_copy(deep=True) for item in value2]
         return None
     if value2 is None:
-        return [item.copy(deep=True) for item in value1]
+        return [item.model_copy(deep=True) for item in value1]
     return combiner(value1, value2)
diff --git a/src/dstack/_internal/server/services/resources.py b/src/dstack/_internal/server/services/resources.py
index 8b38f92f4e..176445d465 100644
--- a/src/dstack/_internal/server/services/resources.py
+++ b/src/dstack/_internal/server/services/resources.py
@@ -1,14 +1,13 @@
 from typing import Optional
 
 import gpuhunt
-from pydantic import parse_obj_as
 
 from dstack._internal.core.models.resources import CPUSpec, ResourcesSpec
 
 
 def set_resources_defaults(resources: ResourcesSpec) -> None:
     # TODO: Remove in 0.20. Use resources.cpu directly
-    cpu = parse_obj_as(CPUSpec, resources.cpu)
+    cpu = CPUSpec.model_validate(resources.cpu)
     if cpu.arch is None:
         gpu = resources.gpu
         if (
diff --git a/src/dstack/_internal/server/services/runner/client.py b/src/dstack/_internal/server/services/runner/client.py
index 1ba5e556ee..695afbca72 100644
--- a/src/dstack/_internal/server/services/runner/client.py
+++ b/src/dstack/_internal/server/services/runner/client.py
@@ -13,7 +13,7 @@
 
 from dstack._internal.core.consts import DSTACK_PROJECT_ENV
 from dstack._internal.core.errors import DstackError
-from dstack._internal.core.models.common import CoreModel, NetworkMode
+from dstack._internal.core.models.common import CoreModel, NetworkMode, validate_extra_ignore
 from dstack._internal.core.models.envs import Env
 from dstack._internal.core.models.instances import GpuDriverInfo
 from dstack._internal.core.models.repos.remote import RemoteRepoCreds
@@ -112,7 +112,7 @@ def get_metrics(self) -> Optional[MetricsResponse]:
         if resp.status_code == 404:
             return None
         resp.raise_for_status()
-        return MetricsResponse.__response__.parse_obj(resp.json())
+        return validate_extra_ignore(MetricsResponse, resp.json())
 
     def submit_job(
         self,
@@ -142,7 +142,7 @@ def submit_job(
                 merged_env.update(router_env)
             if server_access:
                 merged_env.setdefault(DSTACK_PROJECT_ENV, run.project_name)
-            job_spec = job_spec.copy(deep=True)
+            job_spec = job_spec.model_copy(deep=True)
             job_spec.env = merged_env
         quota = server_settings.SERVER_LOG_QUOTA_PER_JOB_HOUR
         body = SubmitBody(
@@ -156,9 +156,8 @@ def submit_job(
             run_spec=run.run_spec,
         )
         resp = self._session.post(
-            # use .json() to encode enums
             self._url("/api/submit"),
-            data=body.json(),
+            data=body.json_for_runner(),
             headers={"Content-Type": "application/json"},
             timeout=REQUEST_TIMEOUT,
         )
@@ -184,14 +183,14 @@ def run_job(self) -> Optional[JobInfoResponse]:
         if not _is_json_response(resp):
             # Old runner or runner failed to get job info
             return None
-        return JobInfoResponse.__response__.parse_obj(resp.json())
+        return validate_extra_ignore(JobInfoResponse, resp.json())
 
     def pull(self, timestamp: int) -> PullResponse:
         resp = self._session.get(
             self._url("/api/pull"), params={"timestamp": timestamp}, timeout=REQUEST_TIMEOUT
         )
         resp.raise_for_status()
-        return PullResponse.__response__.parse_obj(resp.json())
+        return validate_extra_ignore(PullResponse, resp.json())
 
     def stop(self):
         resp = self._session.post(self._url("/api/stop"), timeout=REQUEST_TIMEOUT)
@@ -203,7 +202,7 @@ def _url(self, path: str) -> str:
     def _healthcheck(self) -> HealthcheckResponse:
         resp = self._session.get(self._url("/api/healthcheck"), timeout=REQUEST_TIMEOUT)
         resp.raise_for_status()
-        return HealthcheckResponse.__response__.parse_obj(resp.json())
+        return validate_extra_ignore(HealthcheckResponse, resp.json())
 
     def _negotiate(self, healthcheck_response: Optional[HealthcheckResponse] = None) -> None:
         if healthcheck_response is None:
@@ -638,7 +637,7 @@ def _request(
     ) -> requests.Response:
         url = f"{self._base_url}/{path.lstrip('/')}"
         if body is not None:
-            json = body.dict()
+            json = body.model_dump()
         else:
             json = None
         resp = self._session.request(method, url, json=json, timeout=REQUEST_TIMEOUT)
@@ -649,7 +648,7 @@ def _request(
     _M = TypeVar("_M", bound=CoreModel)
 
     def _response(self, model_cls: type[_M], response: requests.Response) -> _M:
-        return model_cls.__response__.parse_obj(response.json())
+        return validate_extra_ignore(model_cls, response.json())
 
     def _raise_for_status(self, response: requests.Response) -> None:
         try:
@@ -728,7 +727,7 @@ def instance_info_response_to_gpu_driver(
 ) -> Optional[GpuDriverInfo]:
     if response is None or not response.gpu_driver_version:
         return None
-    return GpuDriverInfo.parse_obj(
+    return GpuDriverInfo.model_validate(
         {"vendor": response.gpu_vendor, "version": response.gpu_driver_version}
     )
 
diff --git a/src/dstack/_internal/server/services/runs/__init__.py b/src/dstack/_internal/server/services/runs/__init__.py
index b7668f7d20..02b72c981f 100644
--- a/src/dstack/_internal/server/services/runs/__init__.py
+++ b/src/dstack/_internal/server/services/runs/__init__.py
@@ -18,7 +18,7 @@
     ResourceNotExistsError,
     ServerClientError,
 )
-from dstack._internal.core.models.common import ApplyAction
+from dstack._internal.core.models.common import ApplyAction, validate_json_extra_ignore
 from dstack._internal.core.models.profiles import (
     RetryEvent,
 )
@@ -153,7 +153,7 @@ def get_run_status_change_message(
 
 
 def get_run_spec(run_model: RunModel) -> RunSpec:
-    return RunSpec.__response__.parse_raw(run_model.run_spec)
+    return validate_json_extra_ignore(RunSpec, run_model.run_spec)
 
 
 async def list_user_runs(
@@ -536,14 +536,14 @@ async def get_plan(
     unallocated_resources: bool,
     legacy_repo_dir: bool = False,
 ) -> RunPlan:
-    # Spec must be copied by parsing to calculate merged_profile
-    effective_run_spec = RunSpec.parse_obj(run_spec.dict())
+    effective_run_spec = RunSpec.model_validate(run_spec.model_dump())
     effective_run_spec = await apply_plugin_policies(
         user=user.name,
         project=project.name,
         spec=effective_run_spec,
     )
-    effective_run_spec = RunSpec.parse_obj(effective_run_spec.dict())
+    # Spec must be copied by parsing to calculate merged_profile
+    effective_run_spec = RunSpec.model_validate(effective_run_spec.model_dump())
     validate_run_spec_and_set_defaults(
         user=user,
         run_spec=effective_run_spec,
@@ -603,7 +603,7 @@ async def apply_plan(
         spec=run_spec,
     )
     # Spec must be copied by parsing to calculate merged_profile
-    run_spec = RunSpec.parse_obj(run_spec.dict())
+    run_spec = RunSpec.model_validate(run_spec.model_dump())
     validate_run_spec_and_set_defaults(
         user=user, run_spec=run_spec, legacy_repo_dir=legacy_repo_dir
     )
@@ -658,7 +658,7 @@ async def apply_plan(
         update(RunModel)
         .where(RunModel.id == current_resource.id)
         .values(
-            run_spec=run_spec.json(),
+            run_spec=run_spec.model_dump_json(),
             priority=run_spec.configuration.priority,
             deployment_num=new_deployment_num,
         )
@@ -743,7 +743,7 @@ async def submit_run(
             run_name=run_spec.run_name,
             submitted_at=submitted_at,
             status=initial_status,
-            run_spec=run_spec.json(),
+            run_spec=run_spec.model_dump_json(),
             last_processed_at=submitted_at,
             priority=run_spec.configuration.priority,
             deployment_num=0,
@@ -859,7 +859,7 @@ def create_job_model_for_new_submission(
         last_processed_at=now,
         status=status,
         termination_reason=None,
-        job_spec_data=job.job_spec.json(),
+        job_spec_data=job.job_spec.model_dump_json(),
         job_provisioning_data=None,
         probes=[],
         waiting_master_job=job.job_spec.job_num != 0,
@@ -982,7 +982,7 @@ def run_model_to_run(
 
     service_spec = None
     if run_model.service_spec is not None:
-        service_spec = ServiceSpec.__response__.parse_raw(run_model.service_spec)
+        service_spec = validate_json_extra_ignore(ServiceSpec, run_model.service_spec)
 
     status_message = _get_run_status_message(run_model, job_models=job_models)
     error = _get_run_error(run_model)
diff --git a/src/dstack/_internal/server/services/runs/plan.py b/src/dstack/_internal/server/services/runs/plan.py
index 97d3185088..5da7a7fe77 100644
--- a/src/dstack/_internal/server/services/runs/plan.py
+++ b/src/dstack/_internal/server/services/runs/plan.py
@@ -968,7 +968,7 @@ def _get_backend_offer_identity(offer: InstanceOfferWithAvailability) -> Hashabl
     Needed to deduplicate identical backend offers when merging offers from multiple fleets for
     `dstack offer --fleet ...`.
     """
-    return _freeze_offer_identity_value(offer.dict())
+    return _freeze_offer_identity_value(offer.model_dump())
 
 
 def _freeze_offer_identity_value(value: object) -> Hashable:
diff --git a/src/dstack/_internal/server/services/runs/router_worker_sync.py b/src/dstack/_internal/server/services/runs/router_worker_sync.py
index c9960f54d9..4b9b8af656 100644
--- a/src/dstack/_internal/server/services/runs/router_worker_sync.py
+++ b/src/dstack/_internal/server/services/runs/router_worker_sync.py
@@ -23,6 +23,7 @@
 from typing_extensions import NotRequired
 
 from dstack._internal.core.errors import SSHError
+from dstack._internal.core.models.common import validate_json_extra_ignore
 from dstack._internal.core.models.configurations import ReplicaGroup, ServiceConfiguration
 from dstack._internal.core.models.runs import JobStatus, RunSpec, get_service_port
 from dstack._internal.server.models import JobModel, RunModel
@@ -125,7 +126,7 @@ class _WorkerPayloadResult(TypedDict):
 
 
 def run_model_has_sglang_router_replica_group(run_model: RunModel) -> bool:
-    run_spec = RunSpec.__response__.parse_raw(run_model.run_spec)
+    run_spec = validate_json_extra_ignore(RunSpec, run_model.run_spec)
     return run_spec_has_sglang_router_replica_group(run_spec)
 
 
@@ -626,7 +627,7 @@ async def _build_target_workers(
 
 
 async def sync_router_workers_for_run_model(run_model: RunModel) -> None:
-    run_spec = RunSpec.__response__.parse_raw(run_model.run_spec)
+    run_spec = validate_json_extra_ignore(RunSpec, run_model.run_spec)
     config = run_spec.configuration
     if not isinstance(config, ServiceConfiguration):
         return
diff --git a/src/dstack/_internal/server/services/runs/spec.py b/src/dstack/_internal/server/services/runs/spec.py
index cb989ef5b4..364f81769c 100644
--- a/src/dstack/_internal/server/services/runs/spec.py
+++ b/src/dstack/_internal/server/services/runs/spec.py
@@ -73,6 +73,12 @@ def validate_run_spec_and_set_defaults(
     # If a property is stored in job_spec - resolve the default there.
     # Server defaults are preferable over client defaults so that
     # the defaults depend on the server version, not the client version.
+    #
+    # Callers do not reparse afterwards, so the defaults set here must not touch any field that
+    # `ProfileParams` also declares — `run_spec.merged_profile` is computed at parse time and would
+    # silently keep the pre-default value. The fields written here (`repo_id`, `repo_data`,
+    # `ssh_key_pub`, `configuration.priority`, `configuration.resources`,
+    # `configuration.working_dir`) are none of them declared by `ProfileParams`.
     if run_spec.run_name is not None:
         validate_dstack_resource_name(run_spec.run_name)
     _validate_retry_duration(run_spec)
diff --git a/src/dstack/_internal/server/services/services/__init__.py b/src/dstack/_internal/server/services/services/__init__.py
index b637683af7..df3985b674 100644
--- a/src/dstack/_internal/server/services/services/__init__.py
+++ b/src/dstack/_internal/server/services/services/__init__.py
@@ -93,7 +93,7 @@ async def register_service(session: AsyncSession, run_model: RunModel, run_spec:
             "This dstack-server installation forbids services without a gateway."
             " Please configure a gateway."
         )
-    run_model.service_spec = service_spec.json()
+    run_model.service_spec = service_spec.model_dump_json()
 
 
 async def _register_service_in_gateway(
diff --git a/src/dstack/_internal/server/services/services/options.py b/src/dstack/_internal/server/services/services/options.py
index 3e26be39ba..40b45e797f 100644
--- a/src/dstack/_internal/server/services/services/options.py
+++ b/src/dstack/_internal/server/services/services/options.py
@@ -49,5 +49,5 @@ def get_service_options(conf: ServiceConfiguration) -> dict:
     options = {}
     if conf.model is not None:
         complete_service_model(conf.model, env=conf.env.as_dict())
-        options["openai"] = {"model": conf.model.dict()}
+        options["openai"] = {"model": conf.model.model_dump()}
     return options
diff --git a/src/dstack/_internal/server/services/ssh_fleets/provisioning.py b/src/dstack/_internal/server/services/ssh_fleets/provisioning.py
index 553b82f852..ed48b619e6 100644
--- a/src/dstack/_internal/server/services/ssh_fleets/provisioning.py
+++ b/src/dstack/_internal/server/services/ssh_fleets/provisioning.py
@@ -291,10 +291,10 @@ def host_info_to_instance_type(host_info: Dict[str, Any], arch: GoArchType) -> I
         resources=Resources(
             cpu_arch=arch.to_cpu_architecture(),
             cpus=host_info["cpus"],
-            memory_mib=host_info["memory"] / 1024 / 1024,
+            memory_mib=host_info["memory"] // 1024 // 1024,
             spot=False,
             gpus=gpus,
-            disk=Disk(size_mib=host_info["disk_size"] / 1024 / 1024),
+            disk=Disk(size_mib=host_info["disk_size"] // 1024 // 1024),
         ),
     )
     return instance_type
diff --git a/src/dstack/_internal/server/services/templates.py b/src/dstack/_internal/server/services/templates.py
index 1ac1f357c4..752f45a49e 100644
--- a/src/dstack/_internal/server/services/templates.py
+++ b/src/dstack/_internal/server/services/templates.py
@@ -91,7 +91,7 @@ def _parse_templates(repo_path: Path) -> List[UITemplate]:
             if data.get("type") != "template":
                 logger.debug("Skipping %s: type is not 'template'", entry.name)
                 continue
-            template = UITemplate.parse_obj(data)
+            template = UITemplate.model_validate(data)
             templates.append(template)
         except Exception:
             logger.warning("Skipping invalid template %s", entry.name, exc_info=True)
diff --git a/src/dstack/_internal/server/services/volumes.py b/src/dstack/_internal/server/services/volumes.py
index 9ec85ad8d2..374fe76a4c 100644
--- a/src/dstack/_internal/server/services/volumes.py
+++ b/src/dstack/_internal/server/services/volumes.py
@@ -14,7 +14,8 @@
     ResourceExistsError,
     ServerClientError,
 )
-from dstack._internal.core.models.profiles import parse_duration
+from dstack._internal.core.models.common import validate_json_extra_ignore
+from dstack._internal.core.models.duration import parse_duration
 from dstack._internal.core.models.volumes import (
     AnyVolumeConfiguration,
     Volume,
@@ -300,7 +301,7 @@ async def create_volume(
             user_id=user.id,
             project=project,
             status=VolumeStatus.SUBMITTED,
-            configuration=configuration.json(),
+            configuration=configuration.model_dump_json(),
             auto_cleanup_enabled=_get_autocleanup_enabled(configuration),
             attachments=[],
             created_at=now,
@@ -418,19 +419,21 @@ def volume_model_to_volume(volume_model: VolumeModel) -> Volume:
 
 
 def get_volume_configuration(volume_model: VolumeModel) -> AnyVolumeConfiguration:
-    return VolumeConfiguration.__response__.parse_raw(volume_model.configuration).__root__
+    return validate_json_extra_ignore(VolumeConfiguration, volume_model.configuration).root
 
 
 def get_volume_provisioning_data(volume_model: VolumeModel) -> Optional[VolumeProvisioningData]:
     if volume_model.volume_provisioning_data is None:
         return None
-    return VolumeProvisioningData.__response__.parse_raw(volume_model.volume_provisioning_data)
+    return validate_json_extra_ignore(
+        VolumeProvisioningData, volume_model.volume_provisioning_data
+    )
 
 
 def get_volume_attachment_data(volume_model: VolumeModel) -> Optional[VolumeAttachmentData]:
     if volume_model.volume_attachment_data is None:
         return None
-    return VolumeAttachmentData.__response__.parse_raw(volume_model.volume_attachment_data)
+    return validate_json_extra_ignore(VolumeAttachmentData, volume_model.volume_attachment_data)
 
 
 def get_attachment_data(
@@ -438,7 +441,9 @@ def get_attachment_data(
 ) -> Optional[VolumeAttachmentData]:
     if volume_attachment_model.attachment_data is None:
         return None
-    return VolumeAttachmentData.__response__.parse_raw(volume_attachment_model.attachment_data)
+    return validate_json_extra_ignore(
+        VolumeAttachmentData, volume_attachment_model.attachment_data
+    )
 
 
 def instance_model_to_volume_instance(instance_model: InstanceModel) -> VolumeInstance:
diff --git a/src/dstack/_internal/server/testing/common.py b/src/dstack/_internal/server/testing/common.py
index 9009eec6c4..46b51a189e 100644
--- a/src/dstack/_internal/server/testing/common.py
+++ b/src/dstack/_internal/server/testing/common.py
@@ -26,7 +26,7 @@
     ComputeWithVolumeSupport,
 )
 from dstack._internal.core.models.backends.base import BackendType
-from dstack._internal.core.models.common import NetworkMode
+from dstack._internal.core.models.common import NetworkMode, validate_json_extra_ignore
 from dstack._internal.core.models.compute_groups import (
     ComputeGroupProvisioningData,
     ComputeGroupStatus,
@@ -35,6 +35,7 @@
     AnyRunConfiguration,
     DevEnvironmentConfiguration,
 )
+from dstack._internal.core.models.duration import OptionalIdleDuration
 from dstack._internal.core.models.envs import Env
 from dstack._internal.core.models.fleets import (
     FleetConfiguration,
@@ -408,7 +409,7 @@ async def create_run(
         run_name=run_name,
         status=status,
         termination_reason=termination_reason,
-        run_spec=run_spec.json(),
+        run_spec=run_spec.model_dump_json(),
         last_processed_at=last_processed_at,
         jobs=[],
         priority=priority,
@@ -445,7 +446,7 @@ async def create_job(
 ) -> JobModel:
     if deployment_num is None:
         deployment_num = run.deployment_num
-    run_spec = RunSpec.__response__.parse_raw(run.run_spec)
+    run_spec = validate_json_extra_ignore(RunSpec, run.run_spec)
     job_spec = (
         await get_job_specs_from_run_spec(run_spec=run_spec, secrets={}, replica_num=replica_num)
     )[0]
@@ -464,9 +465,11 @@ async def create_job(
         last_processed_at=last_processed_at,
         status=status,
         termination_reason=termination_reason,
-        job_spec_data=job_spec.json(),
-        job_provisioning_data=job_provisioning_data.json() if job_provisioning_data else None,
-        job_runtime_data=job_runtime_data.json() if job_runtime_data else None,
+        job_spec_data=job_spec.model_dump_json(),
+        job_provisioning_data=job_provisioning_data.model_dump_json()
+        if job_provisioning_data
+        else None,
+        job_runtime_data=job_runtime_data.model_dump_json() if job_runtime_data else None,
         instance=instance,
         instance_assigned=instance_assigned,
         used_instance_id=instance.id if instance is not None else None,
@@ -588,7 +591,7 @@ async def create_compute_group(
         project=project,
         fleet=fleet,
         status=status,
-        provisioning_data=provisioning_data.json(),
+        provisioning_data=provisioning_data.model_dump_json(),
         last_processed_at=last_processed_at,
     )
     session.add(compute_group)
@@ -673,7 +676,7 @@ async def create_gateway(
             domain=wildcard_domain,
             replicas=replicas,
             certificate=certificate,
-        ).json()
+        ).model_dump_json()
     gateway = GatewayModel(
         project_id=project_id,
         backend_id=backend_id,
@@ -732,7 +735,7 @@ async def create_gateway_compute(
             public_ip=True,
             ssh_key_pub=ssh_public_key,
             certificate=None,
-        ).json()
+        ).model_dump_json()
     gateway_compute = GatewayComputeModel(
         gateway_id=gateway_id,
         backend_id=backend_id,
@@ -796,7 +799,7 @@ async def create_fleet(
         name=spec.configuration.name,
         status=status,
         created_at=created_at,
-        spec=spec.json(),
+        spec=spec.model_dump_json(),
         instances=[],
         runs=[],
         last_processed_at=last_processed_at,
@@ -942,17 +945,21 @@ async def create_instance(
         created_at=created_at,
         started_at=created_at,
         finished_at=finished_at,
-        job_provisioning_data=job_provisioning_data.json() if job_provisioning_data else None,
-        offer=offer.json() if offer else None,
+        job_provisioning_data=job_provisioning_data.model_dump_json()
+        if job_provisioning_data
+        else None,
+        offer=offer.model_dump_json() if offer else None,
         price=price,
         region=region,
         backend=backend,
         termination_policy=termination_policy,
         termination_idle_time=termination_idle_time,
-        profile=profile.json(),
-        requirements=requirements.json(),
-        instance_configuration=instance_configuration.json(),
-        remote_connection_info=remote_connection_info.json() if remote_connection_info else None,
+        profile=profile.model_dump_json(),
+        requirements=requirements.model_dump_json(),
+        instance_configuration=instance_configuration.model_dump_json(),
+        remote_connection_info=remote_connection_info.model_dump_json()
+        if remote_connection_info
+        else None,
         volume_attachments=volume_attachments,
         total_blocks=total_blocks,
         busy_blocks=busy_blocks,
@@ -1038,7 +1045,7 @@ def get_remote_connection_info(
     if env is None:
         env = Env()
     elif isinstance(env, dict):
-        env = Env.parse_obj(env)
+        env = Env.model_validate(env)
     return RemoteConnectionInfo(
         host=host,
         port=port,
@@ -1109,8 +1116,8 @@ async def create_volume(
         created_at=created_at,
         last_processed_at=last_processed_at,
         last_job_processed_at=last_job_processed_at,
-        configuration=configuration.json(),
-        volume_provisioning_data=volume_provisioning_data.json()
+        configuration=configuration.model_dump_json(),
+        volume_provisioning_data=volume_provisioning_data.model_dump_json()
         if volume_provisioning_data
         else None,
         attachments=[],
@@ -1170,10 +1177,10 @@ def get_volume_configuration(
     region: str = "eu-west-1",
     size: Optional[Memory] = Memory(100),
     volume_id: Optional[str] = None,
-    auto_cleanup_duration: Optional[Union[str, int]] = None,
+    auto_cleanup_duration: OptionalIdleDuration = None,
 ) -> AnyVolumeConfiguration:
     assert backend != BackendType.KUBERNETES, "use get_kubernetes_volume_configuration() instead"
-    return VolumeConfiguration.parse_obj(
+    return VolumeConfiguration.model_validate(
         dict(
             name=name,
             backend=backend,
@@ -1182,14 +1189,14 @@ def get_volume_configuration(
             volume_id=volume_id,
             auto_cleanup_duration=auto_cleanup_duration,
         )
-    ).__root__
+    ).root
 
 
 def get_kubernetes_volume_configuration(
     name: str = "test-volume",
     size: Optional[Memory] = Memory(100),
     claim_name: Optional[str] = None,
-    auto_cleanup_duration: Optional[Union[str, int]] = None,
+    auto_cleanup_duration: OptionalIdleDuration = None,
     storage_class_name: Optional[str] = None,
 ) -> KubernetesVolumeConfiguration:
     return KubernetesVolumeConfiguration(
@@ -1241,8 +1248,8 @@ async def create_placement_group(
         fleet=fleet,
         name=name,
         created_at=created_at,
-        configuration=configuration.json(),
-        provisioning_data=provisioning_data.json(),
+        configuration=configuration.model_dump_json(),
+        provisioning_data=provisioning_data.model_dump_json(),
         fleet_deleted=fleet_deleted,
         deleted=deleted,
         deleted_at=deleted_at,
diff --git a/src/dstack/_internal/server/utils/routers.py b/src/dstack/_internal/server/utils/routers.py
index e56dcb7be8..4d2eb38a35 100644
--- a/src/dstack/_internal/server/utils/routers.py
+++ b/src/dstack/_internal/server/utils/routers.py
@@ -1,15 +1,17 @@
 from typing import Any, Dict, List, Optional
 
-import orjson
 import packaging.version
 from fastapi import HTTPException, Request, Response, status
 from fastapi.staticfiles import StaticFiles
+from pydantic_core import to_json
 
 from dstack._internal.core.errors import ServerClientError, ServerClientErrorCode
 from dstack._internal.core.models.common import CoreModel
-from dstack._internal.utils.json_utils import get_orjson_default_options, orjson_default
+from dstack._internal.utils.logging import get_logger
 from dstack._internal.utils.version import parse_version
 
+logger = get_logger(__name__)
+
 
 class CustomStaticFiles(StaticFiles):
     """
@@ -27,26 +29,41 @@ async def __call__(self, scope, receive, send) -> None:
         await super().__call__(scope, receive, send)
 
 
-class CustomORJSONResponse(Response):
+class CustomJSONResponse(Response):
     """
-    Custom JSONResponse that uses orjson for serialization.
+    JSONResponse backed by pydantic's own Rust serializer.
 
     It's recommended to return this class from routers directly instead of
     returning pydantic models to avoid the FastAPI's jsonable_encoder overhead.
     See https://fastapi.tiangolo.com/advanced/custom-response/#use-orjsonresponse.
 
     Beware that FastAPI skips model validation when responses are returned directly.
-    If serialization needs to be modified, override `dict()` instead of adding validators.
+    If serialization needs to be modified, add a `@field_serializer`/`@model_serializer`
+    instead of adding validators.
     """
 
     media_type = "application/json"
 
     def render(self, content: Any) -> bytes:
-        return orjson.dumps(
-            content,
-            option=get_orjson_default_options(),
-            default=orjson_default,
-        )
+        # `content` is a model, a list of models, or a plain dict (the `server/compatibility/`
+        # patches mutate models in place, but some routers do assemble dicts), so it has to be
+        # serialized generically rather than through one model's `model_dump_json`.
+        return to_json(content, fallback=_fallback)
+
+
+def _fallback(obj: Any) -> str:
+    """
+    Last resort for a type `to_json` cannot serialize.
+
+    Returning a string keeps one unexpected value from turning the whole response into a 500, but
+    it also puts a `repr` on the wire where the client expects real data, so it must not pass
+    silently: the fix is a `@field_serializer` on the field that produced it.
+    """
+    logger.error(
+        "Response contains a value of non-serializable type %s. Add a serializer for it.",
+        type(obj).__name__,
+    )
+    return str(obj)
 
 
 class BadRequestDetailsModel(CoreModel):
diff --git a/src/dstack/_internal/utils/common.py b/src/dstack/_internal/utils/common.py
index 9c33d2447e..a6c3828a60 100644
--- a/src/dstack/_internal/utils/common.py
+++ b/src/dstack/_internal/utils/common.py
@@ -13,9 +13,10 @@
 from urllib.parse import urlparse
 from uuid import UUID
 
+from pydantic import TypeAdapter
 from typing_extensions import ParamSpec
 
-from dstack._internal.core.models.common import Duration
+from dstack._internal.core.models.duration import Duration
 from dstack._internal.utils.interpolator import InterpolatorError, VariablesInterpolator
 
 
@@ -67,6 +68,23 @@ def get_milliseconds_since_epoch() -> int:
     return int(round(time.time() * 1000))
 
 
+_DATETIME_ADAPTER = TypeAdapter(datetime)
+
+
+def render_datetime_as_api(value: datetime) -> str:
+    """
+    Render a datetime the way the API serializes one.
+
+    Delegates to pydantic rather than post-processing `isoformat()`, so the result cannot drift
+    from what the models emit. The two differ: pydantic v2 spells a zero UTC offset `Z`, while
+    `isoformat()` spells it `+00:00`.
+
+    Only needed where a datetime is formatted by hand. Anything handed to pydantic or to
+    `pydantic_core.to_json` is already rendered this way.
+    """
+    return _DATETIME_ADAPTER.dump_python(value, mode="json")
+
+
 DateFormatter = Callable[[datetime], str]
 
 
diff --git a/src/dstack/_internal/utils/json_schema.py b/src/dstack/_internal/utils/json_schema.py
deleted file mode 100644
index aec695c16d..0000000000
--- a/src/dstack/_internal/utils/json_schema.py
+++ /dev/null
@@ -1,16 +0,0 @@
-def add_extra_schema_types(schema_property: dict, extra_types: list[dict]):
-    if "allOf" in schema_property:
-        refs = [schema_property.pop("allOf")[0]]
-    elif "anyOf" in schema_property:
-        refs = schema_property.pop("anyOf")
-    elif "oneOf" in schema_property:
-        nested = {"oneOf": schema_property.pop("oneOf")}
-        if "discriminator" in schema_property:
-            nested["discriminator"] = schema_property.pop("discriminator")
-        refs = [nested]
-    elif "type" in schema_property:
-        refs = [{"type": schema_property.pop("type")}]
-    else:
-        refs = [{"$ref": schema_property.pop("$ref")}]
-    refs.extend(extra_types)
-    schema_property["anyOf"] = refs
diff --git a/src/dstack/_internal/utils/json_utils.py b/src/dstack/_internal/utils/json_utils.py
deleted file mode 100644
index 9017e94c31..0000000000
--- a/src/dstack/_internal/utils/json_utils.py
+++ /dev/null
@@ -1,54 +0,0 @@
-from typing import Any
-
-import orjson
-from pydantic import BaseModel
-
-FREEZEGUN = True
-try:
-    from freezegun.api import FakeDatetime
-except ImportError:
-    FREEZEGUN = False
-
-
-ASYNCPG = True
-try:
-    import asyncpg.pgproto.pgproto
-except ImportError:
-    ASYNCPG = False
-
-
-def pydantic_orjson_dumps(v: Any, *, default: Any) -> str:
-    return orjson.dumps(
-        v,
-        option=get_orjson_default_options(),
-        default=orjson_default,
-    ).decode()
-
-
-def pydantic_orjson_dumps_with_indent(v: Any, *, default: Any) -> str:
-    return orjson.dumps(
-        v,
-        option=get_orjson_default_options() | orjson.OPT_INDENT_2,
-        default=orjson_default,
-    ).decode()
-
-
-def orjson_default(obj):
-    if isinstance(obj, float):
-        # orjson does not convert float subclasses be default
-        return float(obj)
-    if isinstance(obj, BaseModel):
-        # Allows calling orjson.dumps() on pydantic models
-        # (e.g. to return from the API)
-        return obj.dict()
-    if ASYNCPG:
-        if isinstance(obj, asyncpg.pgproto.pgproto.UUID):
-            return str(obj)
-    if FREEZEGUN:
-        if isinstance(obj, FakeDatetime):
-            return obj.isoformat()
-    raise TypeError
-
-
-def get_orjson_default_options() -> int:
-    return orjson.OPT_NON_STR_KEYS
diff --git a/src/dstack/api/server/__init__.py b/src/dstack/api/server/__init__.py
index d1130713d6..8a2f2a5125 100644
--- a/src/dstack/api/server/__init__.py
+++ b/src/dstack/api/server/__init__.py
@@ -2,7 +2,7 @@
 import os
 import pprint
 import time
-from typing import Dict, List, Optional, Type
+from typing import Dict, List, Optional, Type, Union
 
 import requests
 import requests_unixsocket
@@ -156,7 +156,7 @@ def get_token_hash(self) -> str:
     def _request(
         self,
         path: str,
-        body: Optional[str] = None,
+        body: Optional[Union[str, bytes]] = None,
         raise_for_status: bool = True,
         method: str = "POST",
         **kwargs,
diff --git a/src/dstack/api/server/_auth.py b/src/dstack/api/server/_auth.py
index b944a292a2..e68c38ef04 100644
--- a/src/dstack/api/server/_auth.py
+++ b/src/dstack/api/server/_auth.py
@@ -1,8 +1,7 @@
 from typing import Optional
 
-from pydantic import parse_obj_as
-
 from dstack._internal.core.models.auth import OAuthProviderInfo
+from dstack._internal.core.models.common import validate_extra_ignore
 from dstack._internal.core.models.users import UserWithCreds
 from dstack._internal.server.schemas.auth import (
     OAuthAuthorizeRequest,
@@ -15,16 +14,16 @@
 class AuthAPIClient(APIClientGroup):
     def list_providers(self) -> list[OAuthProviderInfo]:
         resp = self._request("/api/auth/list_providers")
-        return parse_obj_as(list[OAuthProviderInfo.__response__], resp.json())
+        return validate_extra_ignore(list[OAuthProviderInfo], resp.json())
 
     def authorize(self, provider: str, local_port: Optional[int] = None) -> OAuthAuthorizeResponse:
         body = OAuthAuthorizeRequest(local_port=local_port)
-        resp = self._request(f"/api/auth/{provider}/authorize", body=body.json())
-        return parse_obj_as(OAuthAuthorizeResponse.__response__, resp.json())
+        resp = self._request(f"/api/auth/{provider}/authorize", body=body.model_dump_json())
+        return validate_extra_ignore(OAuthAuthorizeResponse, resp.json())
 
     def callback(
         self, provider: str, code: str, state: str, base_url: Optional[str] = None
     ) -> UserWithCreds:
         body = OAuthCallbackRequest(code=code, state=state, base_url=base_url)
-        resp = self._request(f"/api/auth/{provider}/callback", body=body.json())
-        return parse_obj_as(UserWithCreds.__response__, resp.json())
+        resp = self._request(f"/api/auth/{provider}/callback", body=body.model_dump_json())
+        return validate_extra_ignore(UserWithCreds, resp.json())
diff --git a/src/dstack/api/server/_backends.py b/src/dstack/api/server/_backends.py
index 08e5cd597d..4e742f6f7f 100644
--- a/src/dstack/api/server/_backends.py
+++ b/src/dstack/api/server/_backends.py
@@ -1,12 +1,13 @@
 from typing import List
 
-from pydantic import parse_obj_as
+from pydantic import TypeAdapter
 
 from dstack._internal.core.backends.models import (
     AnyBackendConfigWithCreds,
-    AnyBackendConfigWithCredsResponse,
+    AnyBackendConfigWithCredsTagged,
 )
 from dstack._internal.core.models.backends.base import BackendType
+from dstack._internal.core.models.common import validate_extra_ignore
 from dstack._internal.server.schemas.backends import DeleteBackendsRequest
 from dstack.api.server._group import APIClientGroup
 
@@ -15,7 +16,7 @@ class BackendsAPIClient(APIClientGroup):
     def list_backend_types(self) -> List[BackendType]:
         resp = self._request("/api/backends/list_types")
         backend_types = []
-        for value in parse_obj_as(List[str], resp.json()):
+        for value in TypeAdapter(List[str]).validate_python(resp.json()):
             try:
                 backend_types.append(BackendType(value))
             except ValueError:
@@ -25,21 +26,25 @@ def list_backend_types(self) -> List[BackendType]:
     def create(
         self, project_name: str, config: AnyBackendConfigWithCreds
     ) -> AnyBackendConfigWithCreds:
-        resp = self._request(f"/api/project/{project_name}/backends/create", body=config.json())
-        return parse_obj_as(AnyBackendConfigWithCredsResponse, resp.json())
+        resp = self._request(
+            f"/api/project/{project_name}/backends/create", body=config.model_dump_json()
+        )
+        return validate_extra_ignore(AnyBackendConfigWithCredsTagged, resp.json())
 
     def update(
         self, project_name: str, config: AnyBackendConfigWithCreds
     ) -> AnyBackendConfigWithCreds:
-        resp = self._request(f"/api/project/{project_name}/backends/update", body=config.json())
-        return parse_obj_as(AnyBackendConfigWithCredsResponse, resp.json())
+        resp = self._request(
+            f"/api/project/{project_name}/backends/update", body=config.model_dump_json()
+        )
+        return validate_extra_ignore(AnyBackendConfigWithCredsTagged, resp.json())
 
     def delete(self, project_name: str, backends_names: List[BackendType]):
         body = DeleteBackendsRequest(backends_names=backends_names)
-        self._request(f"/api/project/{project_name}/backends/delete", body=body.json())
+        self._request(f"/api/project/{project_name}/backends/delete", body=body.model_dump_json())
 
     def config_info(
         self, project_name: str, backend_name: BackendType
     ) -> AnyBackendConfigWithCreds:
         resp = self._request(f"/api/project/{project_name}/backends/{backend_name}/config_info")
-        return parse_obj_as(AnyBackendConfigWithCredsResponse, resp.json())
+        return validate_extra_ignore(AnyBackendConfigWithCredsTagged, resp.json())
diff --git a/src/dstack/api/server/_events.py b/src/dstack/api/server/_events.py
index 22cd8893cd..132d5994f4 100644
--- a/src/dstack/api/server/_events.py
+++ b/src/dstack/api/server/_events.py
@@ -2,9 +2,8 @@
 from typing import Optional
 from uuid import UUID
 
-from pydantic import parse_obj_as
-
 from dstack._internal.core.compatibility.events import get_list_events_excludes
+from dstack._internal.core.models.common import validate_extra_ignore
 from dstack._internal.core.models.events import Event, EventTargetType
 from dstack._internal.server.schemas.events import LIST_EVENTS_DEFAULT_LIMIT, ListEventsRequest
 from dstack.api.server._group import APIClientGroup
@@ -59,6 +58,6 @@ def list(
             ascending=ascending,
         )
         resp = self._request(
-            "/api/events/list", body=req.json(exclude=get_list_events_excludes(req))
+            "/api/events/list", body=req.model_dump_json(exclude=get_list_events_excludes(req))
         )
-        return parse_obj_as(list[Event.__response__], resp.json())
+        return validate_extra_ignore(list[Event], resp.json())
diff --git a/src/dstack/api/server/_exports.py b/src/dstack/api/server/_exports.py
index f23016011d..7330fbf6a1 100644
--- a/src/dstack/api/server/_exports.py
+++ b/src/dstack/api/server/_exports.py
@@ -1,11 +1,10 @@
 from typing import List
 
-from pydantic import parse_obj_as
-
 from dstack._internal.core.compatibility.exports import (
     get_create_export_excludes,
     get_update_export_excludes,
 )
+from dstack._internal.core.models.common import validate_extra_ignore
 from dstack._internal.core.models.exports import Export
 from dstack._internal.server.schemas.exports import (
     CreateExportRequest,
@@ -18,7 +17,7 @@
 class ExportsAPIClient(APIClientGroup):
     def list(self, project_name: str) -> List[Export]:
         resp = self._request(f"/api/project/{project_name}/exports/list")
-        return parse_obj_as(List[Export.__response__], resp.json())
+        return validate_extra_ignore(List[Export], resp.json())
 
     def create(
         self,
@@ -39,9 +38,9 @@ def create(
         )
         resp = self._request(
             f"/api/project/{project_name}/exports/create",
-            body=body.json(exclude=get_create_export_excludes(body)),
+            body=body.model_dump_json(exclude=get_create_export_excludes(body)),
         )
-        return parse_obj_as(Export.__response__, resp.json())
+        return validate_extra_ignore(Export, resp.json())
 
     def update(
         self,
@@ -70,10 +69,10 @@ def update(
         )
         resp = self._request(
             f"/api/project/{project_name}/exports/update",
-            body=body.json(exclude=get_update_export_excludes(body)),
+            body=body.model_dump_json(exclude=get_update_export_excludes(body)),
         )
-        return parse_obj_as(Export.__response__, resp.json())
+        return validate_extra_ignore(Export, resp.json())
 
     def delete(self, project_name: str, name: str) -> None:
         body = DeleteExportRequest(name=name)
-        self._request(f"/api/project/{project_name}/exports/delete", body=body.json())
+        self._request(f"/api/project/{project_name}/exports/delete", body=body.model_dump_json())
diff --git a/src/dstack/api/server/_files.py b/src/dstack/api/server/_files.py
index e7bdde91a3..d5a7202daf 100644
--- a/src/dstack/api/server/_files.py
+++ b/src/dstack/api/server/_files.py
@@ -1,7 +1,6 @@
 from typing import BinaryIO
 
-from pydantic import parse_obj_as
-
+from dstack._internal.core.models.common import validate_extra_ignore
 from dstack._internal.core.models.files import FileArchive
 from dstack._internal.server.schemas.files import GetFileArchiveByHashRequest
 from dstack.api.server._group import APIClientGroup
@@ -10,9 +9,9 @@
 class FilesAPIClient(APIClientGroup):
     def get_archive_by_hash(self, hash: str) -> FileArchive:
         body = GetFileArchiveByHashRequest(hash=hash)
-        resp = self._request("/api/files/get_archive_by_hash", body=body.json())
-        return parse_obj_as(FileArchive.__response__, resp.json())
+        resp = self._request("/api/files/get_archive_by_hash", body=body.model_dump_json())
+        return validate_extra_ignore(FileArchive, resp.json())
 
     def upload_archive(self, hash: str, fp: BinaryIO) -> FileArchive:
         resp = self._request("/api/files/upload_archive", files={"file": (hash, fp)})
-        return parse_obj_as(FileArchive.__response__, resp.json())
+        return validate_extra_ignore(FileArchive, resp.json())
diff --git a/src/dstack/api/server/_fleets.py b/src/dstack/api/server/_fleets.py
index 93f27e6728..e4780890ec 100644
--- a/src/dstack/api/server/_fleets.py
+++ b/src/dstack/api/server/_fleets.py
@@ -2,14 +2,13 @@
 from typing import List, Optional, Union
 from uuid import UUID
 
-from pydantic import parse_obj_as
-
 from dstack._internal.core.compatibility.fleets import (
     get_apply_plan_excludes,
     get_create_fleet_excludes,
     get_get_plan_excludes,
     patch_fleet_spec,
 )
+from dstack._internal.core.models.common import validate_extra_ignore
 from dstack._internal.core.models.fleets import ApplyFleetPlanInput, Fleet, FleetPlan, FleetSpec
 from dstack._internal.server.schemas.fleets import (
     ApplyFleetPlanRequest,
@@ -26,8 +25,10 @@
 class FleetsAPIClient(APIClientGroup):
     def list(self, project_name: str, *, include_imported: bool = False) -> List[Fleet]:
         body = ListProjectFleetsRequest(include_imported=include_imported)
-        resp = self._request(f"/api/project/{project_name}/fleets/list", body=body.json())
-        return parse_obj_as(List[Fleet.__response__], resp.json())
+        resp = self._request(
+            f"/api/project/{project_name}/fleets/list", body=body.model_dump_json()
+        )
+        return validate_extra_ignore(List[Fleet], resp.json())
 
     def get(
         self, project_name: str, name: Optional[str] = None, fleet_id: Optional[UUID] = None
@@ -39,9 +40,9 @@ def get(
         body = GetFleetRequest(name=name, id=fleet_id)
         resp = self._request(
             f"/api/project/{project_name}/fleets/get",
-            body=body.json(),
+            body=body.model_dump_json(),
         )
-        return parse_obj_as(Fleet.__response__, resp.json())
+        return validate_extra_ignore(Fleet, resp.json())
 
     def get_plan(
         self,
@@ -51,9 +52,9 @@ def get_plan(
         body = GetFleetPlanRequest(spec=spec)
         body = copy.deepcopy(body)
         patch_fleet_spec(body.spec)
-        body_json = body.json(exclude=get_get_plan_excludes(spec))
+        body_json = body.model_dump_json(exclude=get_get_plan_excludes(spec))
         resp = self._request(f"/api/project/{project_name}/fleets/get_plan", body=body_json)
-        return parse_obj_as(FleetPlan.__response__, resp.json())
+        return validate_extra_ignore(FleetPlan, resp.json())
 
     def apply_plan(
         self,
@@ -61,23 +62,25 @@ def apply_plan(
         plan: Union[FleetPlan, ApplyFleetPlanInput],
         force: bool = False,
     ) -> Fleet:
-        plan_input = ApplyFleetPlanInput.__response__.parse_obj(plan)
+        plan_input = validate_extra_ignore(ApplyFleetPlanInput, plan)
         body = ApplyFleetPlanRequest(plan=plan_input, force=force)
         body = copy.deepcopy(body)
         patch_fleet_spec(body.plan.spec)
         if body.plan.current_resource is not None:
             patch_fleet_spec(body.plan.current_resource.spec)
-        body_json = body.json(exclude=get_apply_plan_excludes(plan_input))
+        body_json = body.model_dump_json(exclude=get_apply_plan_excludes(plan_input))
         resp = self._request(f"/api/project/{project_name}/fleets/apply", body=body_json)
-        return parse_obj_as(Fleet.__response__, resp.json())
+        return validate_extra_ignore(Fleet, resp.json())
 
     def delete(self, project_name: str, names: List[str]) -> None:
         body = DeleteFleetsRequest(names=names)
-        self._request(f"/api/project/{project_name}/fleets/delete", body=body.json())
+        self._request(f"/api/project/{project_name}/fleets/delete", body=body.model_dump_json())
 
     def delete_instances(self, project_name: str, name: str, instance_nums: List[int]) -> None:
         body = DeleteFleetInstancesRequest(name=name, instance_nums=instance_nums)
-        self._request(f"/api/project/{project_name}/fleets/delete_instances", body=body.json())
+        self._request(
+            f"/api/project/{project_name}/fleets/delete_instances", body=body.model_dump_json()
+        )
 
     # Deprecated
     # TODO: Remove in 0.21
@@ -89,6 +92,6 @@ def create(
         body = CreateFleetRequest(spec=spec)
         body = copy.deepcopy(body)
         patch_fleet_spec(body.spec)
-        body_json = body.json(exclude=get_create_fleet_excludes(spec))
+        body_json = body.model_dump_json(exclude=get_create_fleet_excludes(spec))
         resp = self._request(f"/api/project/{project_name}/fleets/create", body=body_json)
-        return parse_obj_as(Fleet.__response__, resp.json())
+        return validate_extra_ignore(Fleet, resp.json())
diff --git a/src/dstack/api/server/_gateways.py b/src/dstack/api/server/_gateways.py
index f811894615..715be0f5a2 100644
--- a/src/dstack/api/server/_gateways.py
+++ b/src/dstack/api/server/_gateways.py
@@ -1,11 +1,10 @@
 from typing import List, Optional
 
-from pydantic import parse_obj_as
-
 from dstack._internal.core.compatibility.gateways import (
     get_create_gateway_excludes,
     get_set_default_gateway_excludes,
 )
+from dstack._internal.core.models.common import validate_extra_ignore
 from dstack._internal.core.models.gateways import (
     ApplyGatewayPlanInput,
     Gateway,
@@ -31,25 +30,33 @@ def list(self, project_name: str, *, include_imported: bool = False) -> List[Gat
         body = ListGatewaysRequest(
             include_imported=include_imported,
         )
-        resp = self._request(f"/api/project/{project_name}/gateways/list", body=body.json())
-        return parse_obj_as(List[Gateway.__response__], resp.json())
+        resp = self._request(
+            f"/api/project/{project_name}/gateways/list", body=body.model_dump_json()
+        )
+        return validate_extra_ignore(List[Gateway], resp.json())
 
     def get(self, project_name: str, gateway_name: str) -> Gateway:
         body = GetGatewayRequest(name=gateway_name)
-        resp = self._request(f"/api/project/{project_name}/gateways/get", body=body.json())
-        return parse_obj_as(Gateway.__response__, resp.json())
+        resp = self._request(
+            f"/api/project/{project_name}/gateways/get", body=body.model_dump_json()
+        )
+        return validate_extra_ignore(Gateway, resp.json())
 
     def get_plan(self, project_name: str, spec: GatewaySpec) -> GatewayPlan:
         body = GetGatewayPlanRequest(spec=spec)
-        resp = self._request(f"/api/project/{project_name}/gateways/get_plan", body=body.json())
-        return parse_obj_as(GatewayPlan.__response__, resp.json())
+        resp = self._request(
+            f"/api/project/{project_name}/gateways/get_plan", body=body.model_dump_json()
+        )
+        return validate_extra_ignore(GatewayPlan, resp.json())
 
     def apply_plan(
         self, project_name: str, plan: ApplyGatewayPlanInput, *, force: bool = False
     ) -> Gateway:
         body = ApplyGatewayPlanRequest(plan=plan, force=force)
-        resp = self._request(f"/api/project/{project_name}/gateways/apply", body=body.json())
-        return parse_obj_as(Gateway.__response__, resp.json())
+        resp = self._request(
+            f"/api/project/{project_name}/gateways/apply", body=body.model_dump_json()
+        )
+        return validate_extra_ignore(Gateway, resp.json())
 
     def create(
         self,
@@ -59,13 +66,13 @@ def create(
         body = CreateGatewayRequest(configuration=configuration)
         resp = self._request(
             f"/api/project/{project_name}/gateways/create",
-            body=body.json(exclude=get_create_gateway_excludes(configuration)),
+            body=body.model_dump_json(exclude=get_create_gateway_excludes(configuration)),
         )
-        return parse_obj_as(Gateway.__response__, resp.json())
+        return validate_extra_ignore(Gateway, resp.json())
 
     def delete(self, project_name: str, gateways_names: List[str]) -> None:
         body = DeleteGatewaysRequest(names=gateways_names)
-        self._request(f"/api/project/{project_name}/gateways/delete", body=body.json())
+        self._request(f"/api/project/{project_name}/gateways/delete", body=body.model_dump_json())
 
     def set_default(
         self, project_name: str, gateway_name: str, *, gateway_project: Optional[str] = None
@@ -75,7 +82,7 @@ def set_default(
         body = SetDefaultGatewayRequest(name=gateway_name, gateway_project=gateway_project)
         self._request(
             f"/api/project/{project_name}/gateways/set_default",
-            body=body.json(exclude=get_set_default_gateway_excludes(body)),
+            body=body.model_dump_json(exclude=get_set_default_gateway_excludes(body)),
         )
 
     def set_wildcard_domain(
@@ -83,6 +90,7 @@ def set_wildcard_domain(
     ) -> Gateway:
         body = SetWildcardDomainRequest(name=gateway_name, wildcard_domain=wildcard_domain)
         resp = self._request(
-            f"/api/project/{project_name}/gateways/set_wildcard_domain", body=body.json()
+            f"/api/project/{project_name}/gateways/set_wildcard_domain",
+            body=body.model_dump_json(),
         )
-        return parse_obj_as(Gateway.__response__, resp.json())
+        return validate_extra_ignore(Gateway, resp.json())
diff --git a/src/dstack/api/server/_gpus.py b/src/dstack/api/server/_gpus.py
index 78886d4d4e..9939eda4f0 100644
--- a/src/dstack/api/server/_gpus.py
+++ b/src/dstack/api/server/_gpus.py
@@ -1,8 +1,7 @@
 from typing import List, Literal, Optional, cast
 
-from pydantic import parse_obj_as
-
 from dstack._internal.core.compatibility.gpus import get_list_gpus_excludes
+from dstack._internal.core.models.common import validate_extra_ignore
 from dstack._internal.core.models.gpus import GpuGroup
 from dstack._internal.core.models.runs import RunSpec
 from dstack._internal.server.schemas.gpus import ListGpusRequest, ListGpusResponse
@@ -26,6 +25,6 @@ def list_gpus(
         )
         resp = self._request(
             f"/api/project/{project_name}/gpus/list",
-            body=body.json(exclude=get_list_gpus_excludes(body)),
+            body=body.model_dump_json(exclude=get_list_gpus_excludes(body)),
         )
-        return parse_obj_as(ListGpusResponse.__response__, resp.json()).gpus
+        return validate_extra_ignore(ListGpusResponse, resp.json()).gpus
diff --git a/src/dstack/api/server/_group.py b/src/dstack/api/server/_group.py
index 9d3ec1918a..fae9dfcc87 100644
--- a/src/dstack/api/server/_group.py
+++ b/src/dstack/api/server/_group.py
@@ -1,5 +1,5 @@
 from logging import Logger
-from typing import Optional
+from typing import Optional, Union
 
 import requests
 from typing_extensions import Protocol
@@ -9,7 +9,7 @@ class APIRequest(Protocol):
     def __call__(
         self,
         path: str,
-        body: Optional[str] = None,
+        body: Optional[Union[str, bytes]] = None,
         raise_for_status: bool = True,
         method: str = "POST",
         **kwargs,
diff --git a/src/dstack/api/server/_imports.py b/src/dstack/api/server/_imports.py
index bcc1abb162..4fc712bf60 100644
--- a/src/dstack/api/server/_imports.py
+++ b/src/dstack/api/server/_imports.py
@@ -1,7 +1,6 @@
 from typing import List
 
-from pydantic import parse_obj_as
-
+from dstack._internal.core.models.common import validate_extra_ignore
 from dstack._internal.core.models.imports import Import
 from dstack._internal.server.schemas.imports import DeleteImportRequest
 from dstack.api.server._group import APIClientGroup
@@ -10,10 +9,10 @@
 class ImportsAPIClient(APIClientGroup):
     def list(self, project_name: str) -> List[Import]:
         resp = self._request(f"/api/project/{project_name}/imports/list")
-        return parse_obj_as(List[Import.__response__], resp.json())
+        return validate_extra_ignore(List[Import], resp.json())
 
     def delete(self, *, project_name: str, export_project_name: str, export_name: str) -> None:
         body = DeleteImportRequest(
             export_project_name=export_project_name, export_name=export_name
         )
-        self._request(f"/api/project/{project_name}/imports/delete", body=body.json())
+        self._request(f"/api/project/{project_name}/imports/delete", body=body.model_dump_json())
diff --git a/src/dstack/api/server/_logs.py b/src/dstack/api/server/_logs.py
index 7cdfc246f7..7a709dbc6e 100644
--- a/src/dstack/api/server/_logs.py
+++ b/src/dstack/api/server/_logs.py
@@ -1,6 +1,5 @@
-from pydantic import parse_obj_as
-
 from dstack._internal.core.compatibility.logs import get_poll_logs_excludes
+from dstack._internal.core.models.common import validate_extra_ignore
 from dstack._internal.core.models.logs import JobSubmissionLogs
 from dstack._internal.server.schemas.logs import PollLogsRequest
 from dstack.api.server._group import APIClientGroup
@@ -10,6 +9,6 @@ class LogsAPIClient(APIClientGroup):
     def poll(self, project_name: str, body: PollLogsRequest) -> JobSubmissionLogs:
         resp = self._request(
             f"/api/project/{project_name}/logs/poll",
-            body=body.json(exclude=get_poll_logs_excludes(body)),
+            body=body.model_dump_json(exclude=get_poll_logs_excludes(body)),
         )
-        return parse_obj_as(JobSubmissionLogs.__response__, resp.json())
+        return validate_extra_ignore(JobSubmissionLogs, resp.json())
diff --git a/src/dstack/api/server/_metrics.py b/src/dstack/api/server/_metrics.py
index 8b378c89b3..b9604461db 100644
--- a/src/dstack/api/server/_metrics.py
+++ b/src/dstack/api/server/_metrics.py
@@ -1,5 +1,4 @@
-from pydantic import parse_obj_as
-
+from dstack._internal.core.models.common import validate_extra_ignore
 from dstack._internal.core.models.metrics import JobMetrics
 from dstack.api.server._group import APIClientGroup
 
@@ -20,4 +19,4 @@ def get_job_metrics(
                 "job_num": job_num,
             },
         )
-        return parse_obj_as(JobMetrics.__response__, resp.json())
+        return validate_extra_ignore(JobMetrics, resp.json())
diff --git a/src/dstack/api/server/_projects.py b/src/dstack/api/server/_projects.py
index 6c45f63364..8feff6630f 100644
--- a/src/dstack/api/server/_projects.py
+++ b/src/dstack/api/server/_projects.py
@@ -1,10 +1,10 @@
-import json
 from datetime import datetime
 from typing import Any, List, Literal, Optional, Union, overload
 from uuid import UUID
 
-from pydantic import parse_obj_as
+from pydantic_core import to_json
 
+from dstack._internal.core.models.common import validate_extra_ignore
 from dstack._internal.core.models.projects import (
     Project,
     ProjectsInfoList,
@@ -71,54 +71,64 @@ def list(
         if name_pattern is not None:
             body["name_pattern"] = name_pattern
         if prev_created_at is not None:
-            body["prev_created_at"] = prev_created_at.isoformat()
+            body["prev_created_at"] = prev_created_at
         if prev_id is not None:
-            body["prev_id"] = str(prev_id)
+            body["prev_id"] = prev_id
         if limit is not None:
             body["limit"] = limit
         if ascending is not None:
             body["ascending"] = ascending
-        resp = self._request("/api/projects/list", body=json.dumps(body))
+        resp = self._request("/api/projects/list", body=to_json(body))
         resp_json = resp.json()
         if isinstance(resp_json, list):
-            return parse_obj_as(List[Project.__response__], resp_json)
-        return parse_obj_as(ProjectsInfoList.__response__, resp_json)
+            return validate_extra_ignore(List[Project], resp_json)
+        return validate_extra_ignore(ProjectsInfoList, resp_json)
 
     def create(self, project_name: str, is_public: bool = False) -> Project:
         body = CreateProjectRequest(project_name=project_name, is_public=is_public)
-        resp = self._request("/api/projects/create", body=body.json())
-        return parse_obj_as(Project.__response__, resp.json())
+        resp = self._request("/api/projects/create", body=body.model_dump_json())
+        return validate_extra_ignore(Project, resp.json())
 
     def delete(self, projects_names: List[str]):
         body = DeleteProjectsRequest(projects_names=projects_names)
-        self._request("/api/projects/delete", body=body.json())
+        self._request("/api/projects/delete", body=body.model_dump_json())
 
     def get(self, project_name: str) -> Project:
         resp = self._request(f"/api/projects/{project_name}/get")
-        return parse_obj_as(Project.__response__, resp.json())
+        return validate_extra_ignore(Project, resp.json())
 
     def set_members(self, project_name: str, members: List[MemberSetting]) -> Project:
         body = SetProjectMembersRequest(members=members)
-        resp = self._request(f"/api/projects/{project_name}/set_members", body=body.json())
-        return parse_obj_as(Project.__response__, resp.json())
+        resp = self._request(
+            f"/api/projects/{project_name}/set_members", body=body.model_dump_json()
+        )
+        return validate_extra_ignore(Project, resp.json())
 
     def add_member(self, project_name: str, username: str, project_role: ProjectRole) -> Project:
         member_setting = MemberSetting(username=username, project_role=project_role)
         body = AddProjectMemberRequest(members=[member_setting])
-        resp = self._request(f"/api/projects/{project_name}/add_members", body=body.json())
-        return parse_obj_as(Project.__response__, resp.json())
+        resp = self._request(
+            f"/api/projects/{project_name}/add_members", body=body.model_dump_json()
+        )
+        return validate_extra_ignore(Project, resp.json())
 
     def add_members(self, project_name: str, members: List[MemberSetting]) -> Project:
         body = AddProjectMemberRequest(members=members)
-        resp = self._request(f"/api/projects/{project_name}/add_members", body=body.json())
-        return parse_obj_as(Project.__response__, resp.json())
+        resp = self._request(
+            f"/api/projects/{project_name}/add_members", body=body.model_dump_json()
+        )
+        return validate_extra_ignore(Project, resp.json())
 
     def remove_member(self, project_name: str, username: str) -> Project:
         body = RemoveProjectMemberRequest(usernames=[username])
-        resp = self._request(f"/api/projects/{project_name}/remove_members", body=body.json())
-        return parse_obj_as(Project.__response__, resp.json())
+        resp = self._request(
+            f"/api/projects/{project_name}/remove_members", body=body.model_dump_json()
+        )
+        return validate_extra_ignore(Project, resp.json())
 
     def remove_members(self, project_name: str, usernames: List[str]) -> Project:
         body = RemoveProjectMemberRequest(usernames=usernames)
-        resp = self._request(f"/api/projects/{project_name}/remove_members", body=body.json())
-        return parse_obj_as(Project.__response__, resp.json())
+        resp = self._request(
+            f"/api/projects/{project_name}/remove_members", body=body.model_dump_json()
+        )
+        return validate_extra_ignore(Project, resp.json())
diff --git a/src/dstack/api/server/_repos.py b/src/dstack/api/server/_repos.py
index 03f9eb9eab..89a34da5ee 100644
--- a/src/dstack/api/server/_repos.py
+++ b/src/dstack/api/server/_repos.py
@@ -1,7 +1,6 @@
 from typing import BinaryIO, List, Optional
 
-from pydantic import parse_obj_as
-
+from dstack._internal.core.models.common import validate_extra_ignore
 from dstack._internal.core.models.repos import (
     AnyRepoInfo,
     RemoteRepoCreds,
@@ -19,7 +18,7 @@
 class ReposAPIClient(APIClientGroup):
     def list(self, project_name: str) -> List[RepoHead]:
         resp = self._request(f"/api/project/{project_name}/repos/list")
-        return parse_obj_as(List[RepoHead.__response__], resp.json())
+        return validate_extra_ignore(List[RepoHead], resp.json())
 
     def get(
         self, project_name: str, repo_id: str, include_creds: Optional[bool] = None
@@ -30,13 +29,13 @@ def get(
                 " the repo without creds. Use `get_with_creds()` to get the repo with creds"
             )
         body = GetRepoRequest(repo_id=repo_id, include_creds=False)
-        resp = self._request(f"/api/project/{project_name}/repos/get", body=body.json())
-        return parse_obj_as(RepoHead.__response__, resp.json())
+        resp = self._request(f"/api/project/{project_name}/repos/get", body=body.model_dump_json())
+        return validate_extra_ignore(RepoHead, resp.json())
 
     def get_with_creds(self, project_name: str, repo_id: str) -> RepoHeadWithCreds:
         body = GetRepoRequest(repo_id=repo_id, include_creds=True)
-        resp = self._request(f"/api/project/{project_name}/repos/get", body=body.json())
-        return parse_obj_as(RepoHeadWithCreds.__response__, resp.json())
+        resp = self._request(f"/api/project/{project_name}/repos/get", body=body.model_dump_json())
+        return validate_extra_ignore(RepoHeadWithCreds, resp.json())
 
     def init(
         self,
@@ -50,11 +49,11 @@ def init(
             repo_info=repo_info,
             repo_creds=repo_creds,
         )
-        self._request(f"/api/project/{project_name}/repos/init", body=body.json())
+        self._request(f"/api/project/{project_name}/repos/init", body=body.model_dump_json())
 
     def delete(self, project_name: str, repos_ids: List[str]):
         body = DeleteReposRequest(repos_ids=repos_ids)
-        self._request(f"/api/project/{project_name}/repos/delete", body=body.json())
+        self._request(f"/api/project/{project_name}/repos/delete", body=body.model_dump_json())
 
     def upload_code(self, project_name: str, repo_id: str, code_hash: str, fp: BinaryIO):
         self._request(
diff --git a/src/dstack/api/server/_runs.py b/src/dstack/api/server/_runs.py
index e0e0bdd48a..41d5e096af 100644
--- a/src/dstack/api/server/_runs.py
+++ b/src/dstack/api/server/_runs.py
@@ -3,14 +3,13 @@
 from typing import List, Optional, Union
 from uuid import UUID
 
-from pydantic import parse_obj_as
-
 from dstack._internal.core.compatibility.runs import (
     get_apply_plan_excludes,
     get_get_plan_excludes,
     get_list_runs_excludes,
     patch_run_spec,
 )
+from dstack._internal.core.models.common import validate_extra_ignore
 from dstack._internal.core.models.runs import (
     ApplyRunPlanInput,
     Run,
@@ -55,9 +54,9 @@ def list(
             ascending=ascending,
         )
         resp = self._request(
-            "/api/runs/list", body=body.json(exclude=get_list_runs_excludes(body))
+            "/api/runs/list", body=body.model_dump_json(exclude=get_list_runs_excludes(body))
         )
-        return parse_obj_as(List[Run.__response__], resp.json())
+        return validate_extra_ignore(List[Run], resp.json())
 
     def get(
         self, project_name: str, run_name: Optional[str] = None, run_id: Optional[UUID] = None
@@ -67,9 +66,9 @@ def get(
         if run_name is not None and run_id is not None:
             raise ValueError("Cannot specify both run_name and run_id")
         body = GetRunRequest(run_name=run_name, id=run_id)
-        json_body = body.json()
+        json_body = body.model_dump_json()
         resp = self._request(f"/api/project/{project_name}/runs/get", body=json_body)
-        return parse_obj_as(Run.__response__, resp.json())
+        return validate_extra_ignore(Run, resp.json())
 
     def get_plan(
         self,
@@ -89,9 +88,9 @@ def get_plan(
         patch_run_spec(body.run_spec)
         resp = self._request(
             f"/api/project/{project_name}/runs/get_plan",
-            body=body.json(exclude=get_get_plan_excludes(body)),
+            body=body.model_dump_json(exclude=get_get_plan_excludes(body)),
         )
-        return parse_obj_as(RunPlan.__response__, resp.json())
+        return validate_extra_ignore(RunPlan, resp.json())
 
     def apply_plan(
         self,
@@ -99,7 +98,7 @@ def apply_plan(
         plan: Union[RunPlan, ApplyRunPlanInput],
         force: bool = False,
     ) -> Run:
-        plan_input: ApplyRunPlanInput = ApplyRunPlanInput.__response__.parse_obj(plan)
+        plan_input: ApplyRunPlanInput = validate_extra_ignore(ApplyRunPlanInput, plan)
         body = ApplyRunPlanRequest(plan=plan_input, force=force)
         body = copy.deepcopy(body)
         patch_run_spec(body.plan.run_spec)
@@ -107,14 +106,14 @@ def apply_plan(
             patch_run_spec(body.plan.current_resource.run_spec)
         resp = self._request(
             f"/api/project/{project_name}/runs/apply",
-            body=body.json(exclude=get_apply_plan_excludes(plan_input)),
+            body=body.model_dump_json(exclude=get_apply_plan_excludes(plan_input)),
         )
-        return parse_obj_as(Run.__response__, resp.json())
+        return validate_extra_ignore(Run, resp.json())
 
     def stop(self, project_name: str, runs_names: List[str], abort: bool):
         body = StopRunsRequest(runs_names=runs_names, abort=abort)
-        self._request(f"/api/project/{project_name}/runs/stop", body=body.json())
+        self._request(f"/api/project/{project_name}/runs/stop", body=body.model_dump_json())
 
     def delete(self, project_name: str, runs_names: List[str]):
         body = DeleteRunsRequest(runs_names=runs_names)
-        self._request(f"/api/project/{project_name}/runs/delete", body=body.json())
+        self._request(f"/api/project/{project_name}/runs/delete", body=body.model_dump_json())
diff --git a/src/dstack/api/server/_secrets.py b/src/dstack/api/server/_secrets.py
index 1efcdac36e..2c0d050d27 100644
--- a/src/dstack/api/server/_secrets.py
+++ b/src/dstack/api/server/_secrets.py
@@ -1,7 +1,6 @@
 from typing import List
 
-from pydantic import parse_obj_as
-
+from dstack._internal.core.models.common import validate_extra_ignore
 from dstack._internal.core.models.secrets import Secret
 from dstack._internal.server.schemas.secrets import (
     CreateOrUpdateSecretRequest,
@@ -14,12 +13,14 @@
 class SecretsAPIClient(APIClientGroup):
     def list(self, project_name: str) -> List[Secret]:
         resp = self._request(f"/api/project/{project_name}/secrets/list")
-        return parse_obj_as(List[Secret.__response__], resp.json())
+        return validate_extra_ignore(List[Secret], resp.json())
 
     def get(self, project_name: str, name: str) -> Secret:
         body = GetSecretRequest(name=name)
-        resp = self._request(f"/api/project/{project_name}/secrets/get", body=body.json())
-        return parse_obj_as(Secret.__response__, resp.json())
+        resp = self._request(
+            f"/api/project/{project_name}/secrets/get", body=body.model_dump_json()
+        )
+        return validate_extra_ignore(Secret, resp.json())
 
     def create_or_update(self, project_name: str, name: str, value: str) -> Secret:
         body = CreateOrUpdateSecretRequest(
@@ -27,10 +28,10 @@ def create_or_update(self, project_name: str, name: str, value: str) -> Secret:
             value=value,
         )
         resp = self._request(
-            f"/api/project/{project_name}/secrets/create_or_update", body=body.json()
+            f"/api/project/{project_name}/secrets/create_or_update", body=body.model_dump_json()
         )
-        return parse_obj_as(Secret.__response__, resp.json())
+        return validate_extra_ignore(Secret, resp.json())
 
     def delete(self, project_name: str, names: List[str]):
         body = DeleteSecretsRequest(secrets_names=names)
-        self._request(f"/api/project/{project_name}/secrets/delete", body=body.json())
+        self._request(f"/api/project/{project_name}/secrets/delete", body=body.model_dump_json())
diff --git a/src/dstack/api/server/_users.py b/src/dstack/api/server/_users.py
index ff1bab3d48..91ac218536 100644
--- a/src/dstack/api/server/_users.py
+++ b/src/dstack/api/server/_users.py
@@ -1,11 +1,10 @@
-import json
 from datetime import datetime
 from typing import Any, List, Optional
 from uuid import UUID
 
-from pydantic import parse_obj_as
-from pydantic.json import pydantic_encoder
+from pydantic_core import to_json
 
+from dstack._internal.core.models.common import validate_extra_ignore
 from dstack._internal.core.models.users import (
     GlobalRole,
     User,
@@ -47,36 +46,34 @@ def list(
         if ascending is not None:
             body["ascending"] = ascending
         if body:
-            resp = self._request(
-                "/api/users/list", body=json.dumps(body, default=pydantic_encoder)
-            )
+            resp = self._request("/api/users/list", body=to_json(body))
         else:
             resp = self._request("/api/users/list")
         resp_json = resp.json()
         if isinstance(resp_json, list):
-            return parse_obj_as(List[User.__response__], resp_json)
-        return parse_obj_as(UsersInfoList.__response__, resp_json)
+            return validate_extra_ignore(List[User], resp_json)
+        return validate_extra_ignore(UsersInfoList, resp_json)
 
     def get_my_user(self) -> UserWithCreds:
         resp = self._request("/api/users/get_my_user")
-        return parse_obj_as(UserWithCreds.__response__, resp.json())
+        return validate_extra_ignore(UserWithCreds, resp.json())
 
     def get_user(self, username: str) -> User:
         body = GetUserRequest(username=username)
-        resp = self._request("/api/users/get_user", body=body.json())
-        return parse_obj_as(User.__response__, resp.json())
+        resp = self._request("/api/users/get_user", body=body.model_dump_json())
+        return validate_extra_ignore(User, resp.json())
 
     def create(self, username: str, global_role: GlobalRole) -> User:
         body = CreateUserRequest(username=username, global_role=global_role, email=None)
-        resp = self._request("/api/users/create", body=body.json())
-        return parse_obj_as(User.__response__, resp.json())
+        resp = self._request("/api/users/create", body=body.model_dump_json())
+        return validate_extra_ignore(User, resp.json())
 
     def update(self, username: str, global_role: GlobalRole) -> User:
         body = UpdateUserRequest(username=username, global_role=global_role, email=None)
-        resp = self._request("/api/users/update", body=body.json())
-        return parse_obj_as(User.__response__, resp.json())
+        resp = self._request("/api/users/update", body=body.model_dump_json())
+        return validate_extra_ignore(User, resp.json())
 
     def refresh_token(self, username: str) -> UserWithCreds:
         body = RefreshTokenRequest(username=username)
-        resp = self._request("/api/users/refresh_token", body=body.json())
-        return parse_obj_as(UserWithCreds.__response__, resp.json())
+        resp = self._request("/api/users/refresh_token", body=body.model_dump_json())
+        return validate_extra_ignore(UserWithCreds, resp.json())
diff --git a/src/dstack/api/server/_volumes.py b/src/dstack/api/server/_volumes.py
index 5cf56afc3d..6187118a64 100644
--- a/src/dstack/api/server/_volumes.py
+++ b/src/dstack/api/server/_volumes.py
@@ -1,8 +1,7 @@
 from typing import List
 
-from pydantic import parse_obj_as
-
 from dstack._internal.core.compatibility.volumes import get_create_volume_excludes
+from dstack._internal.core.models.common import validate_extra_ignore
 from dstack._internal.core.models.volumes import AnyVolumeConfiguration, Volume
 from dstack._internal.server.schemas.volumes import (
     CreateVolumeRequest,
@@ -15,12 +14,14 @@
 class VolumesAPIClient(APIClientGroup):
     def list(self, project_name: str) -> List[Volume]:
         resp = self._request(f"/api/project/{project_name}/volumes/list")
-        return parse_obj_as(List[Volume.__response__], resp.json())
+        return validate_extra_ignore(List[Volume], resp.json())
 
     def get(self, project_name: str, name: str) -> Volume:
         body = GetVolumeRequest(name=name)
-        resp = self._request(f"/api/project/{project_name}/volumes/get", body=body.json())
-        return parse_obj_as(Volume.__response__, resp.json())
+        resp = self._request(
+            f"/api/project/{project_name}/volumes/get", body=body.model_dump_json()
+        )
+        return validate_extra_ignore(Volume, resp.json())
 
     def create(
         self,
@@ -30,10 +31,10 @@ def create(
         body = CreateVolumeRequest(configuration=configuration)
         resp = self._request(
             f"/api/project/{project_name}/volumes/create",
-            body=body.json(exclude=get_create_volume_excludes(configuration)),
+            body=body.model_dump_json(exclude=get_create_volume_excludes(configuration)),
         )
-        return parse_obj_as(Volume.__response__, resp.json())
+        return validate_extra_ignore(Volume, resp.json())
 
     def delete(self, project_name: str, names: List[str]) -> None:
         body = DeleteVolumesRequest(names=names)
-        self._request(f"/api/project/{project_name}/volumes/delete", body=body.json())
+        self._request(f"/api/project/{project_name}/volumes/delete", body=body.model_dump_json())
diff --git a/src/dstack/api/utils.py b/src/dstack/api/utils.py
index 66bec859fb..1c559aab2c 100644
--- a/src/dstack/api/utils.py
+++ b/src/dstack/api/utils.py
@@ -90,7 +90,7 @@ def _load_profile_from_path(profiles_path: Path, profile_name: Optional[str]) ->
 
     try:
         with profiles_path.open("r") as f:
-            config = ProfilesConfig.parse_obj(yaml.safe_load(f))
+            config = ProfilesConfig.model_validate(yaml.safe_load(f))
     except FileNotFoundError:
         return None
     except ValidationError as e:
diff --git a/src/dstack/plugins/builtin/rest_plugin/_models.py b/src/dstack/plugins/builtin/rest_plugin/_models.py
index ee3a042464..28887bf86f 100644
--- a/src/dstack/plugins/builtin/rest_plugin/_models.py
+++ b/src/dstack/plugins/builtin/rest_plugin/_models.py
@@ -16,13 +16,6 @@ class SpecApplyRequest(BaseModel, Generic[SpecType]):
     project: Annotated[str, Field(description="The name of the project the request is for")]
     spec: Annotated[SpecType, Field(description="The spec to be applied")]
 
-    # Override dict() to remove __orig_class__ attribute and avoid "TypeError: Object of type _GenericAlias is not JSON serializable"
-    # error. This issue doesn't happen though when running the code in pytest, only when running the server.
-    def dict(self, *args, **kwargs):
-        d = super().dict(*args, **kwargs)
-        d.pop("__orig_class__", None)
-        return d
-
 
 RunSpecRequest = SpecApplyRequest[RunSpec]
 FleetSpecRequest = SpecApplyRequest[FleetSpec]
diff --git a/src/dstack/plugins/builtin/rest_plugin/_plugin.py b/src/dstack/plugins/builtin/rest_plugin/_plugin.py
index 210dd50e19..a4d796fd1a 100644
--- a/src/dstack/plugins/builtin/rest_plugin/_plugin.py
+++ b/src/dstack/plugins/builtin/rest_plugin/_plugin.py
@@ -55,10 +55,13 @@ def _call_plugin_service(
         excludes: Optional[Dict],
     ) -> ApplySpec:
         response = None
+        # `{"spec": None}` is not a valid `exclude` value in pydantic v2 (it wants a set, a nested
+        # mapping, or a bool), so omit the key entirely when there is nothing to exclude.
+        exclude = {"spec": excludes} if excludes is not None else None
         try:
             response = requests.post(
                 f"{self._plugin_service_uri}{endpoint}",
-                json=spec_request.dict(exclude={"spec": excludes}),
+                json=spec_request.model_dump(exclude=exclude),
                 headers={"accept": "application/json", "Content-Type": "application/json"},
                 timeout=PLUGIN_REQUEST_TIMEOUT_SEC,
             )
diff --git a/src/tests/_internal/cli/commands/test_preset.py b/src/tests/_internal/cli/commands/test_preset.py
index 5f8c17c654..84b88ae3d0 100644
--- a/src/tests/_internal/cli/commands/test_preset.py
+++ b/src/tests/_internal/cli/commands/test_preset.py
@@ -8,6 +8,7 @@
 
 from dstack._internal.cli.services.presets import output as presets_utils
 from dstack._internal.cli.services.presets.store import PresetStore
+from dstack._internal.utils.common import render_datetime_as_api
 from tests._internal.cli.common import plain_console, run_dstack_cli
 from tests._internal.cli.preset_factories import get_preset
 
@@ -162,7 +163,7 @@ def test_gets_complete_preset_as_json_without_api_client(self, tmp_path, capsys)
 
         data = json.loads(capsys.readouterr().out)
         assert data["id"] == preset.id
-        assert data["created_at"] == preset.created_at.isoformat()
+        assert data["created_at"] == render_datetime_as_api(preset.created_at)
         assert data["context_length"] == 32768
         assert data["validations"][0]["benchmark"]["metrics"]["total_output_tokens"] == 2048
 
@@ -183,7 +184,7 @@ def test_lists_complete_presets_as_json(self, tmp_path, capsys, args):
         assert len(output["presets"]) == 1
         data = output["presets"][0]
         assert data["id"] == preset.id
-        assert data["created_at"] == preset.created_at.isoformat()
+        assert data["created_at"] == render_datetime_as_api(preset.created_at)
         assert data["context_length"] == 32768
         assert data["validations"][0]["benchmark"]["metrics"]["total_output_tokens"] == 2048
 
@@ -195,10 +196,12 @@ def test_deletes_all_presets_of_model_keeping_others_without_api_client(
         preset = get_preset()
         store = PresetStore(tmp_path / ".dstack" / "presets")
         store.save(preset)
-        store.save(preset.copy(update={"id": "01234567"}))
+        store.save(preset.model_copy(update={"id": "01234567"}))
         # A preset of a different model must survive the delete.
         store.save(
-            preset.copy(update={"id": "89abcdef", "base": "meta/Llama-4", "model": "meta/Llama-4"})
+            preset.model_copy(
+                update={"id": "89abcdef", "base": "meta/Llama-4", "model": "meta/Llama-4"}
+            )
         )
 
         with patch("dstack.api.Client.from_config") as from_config:
@@ -220,7 +223,9 @@ def test_lists_presets_filtered_by_model(self, tmp_path, capsys, flag_attribute)
         store = PresetStore(tmp_path / ".dstack" / "presets")
         store.save(preset)
         store.save(
-            preset.copy(update={"id": "01234567", "base": "meta/Llama-4", "model": "meta/Llama-4"})
+            preset.model_copy(
+                update={"id": "01234567", "base": "meta/Llama-4", "model": "meta/Llama-4"}
+            )
         )
 
         args = ["preset", "list", "--json", flag, getattr(preset, attribute)]
@@ -368,7 +373,7 @@ def test_apply_requires_preset_id(self, tmp_path, capsys):
 
 class TestPresetNameClaims:
     def test_create_detaches_the_name_from_the_old_preset(self, tmp_path):
-        preset = get_preset().copy(update={"name": "qwen"})
+        preset = get_preset().model_copy(update={"name": "qwen"})
         store = PresetStore(tmp_path / ".dstack" / "presets")
         store.save(preset)
         configuration_path = tmp_path / "preset.dstack.yml"
@@ -391,7 +396,7 @@ def test_create_detaches_the_name_from_the_old_preset(self, tmp_path):
         assert store.get(preset.id).name is None
 
     def test_create_without_confirmation_exits_before_creating(self, tmp_path):
-        preset = get_preset().copy(update={"name": "qwen"})
+        preset = get_preset().model_copy(update={"name": "qwen"})
         store = PresetStore(tmp_path / ".dstack" / "presets")
         store.save(preset)
         configuration_path = tmp_path / "preset.dstack.yml"
@@ -414,7 +419,7 @@ def test_create_without_confirmation_exits_before_creating(self, tmp_path):
         assert store.get(preset.id).name == "qwen"
 
     def test_get_and_delete_resolve_names(self, tmp_path, capsys):
-        preset = get_preset().copy(update={"name": "qwen"})
+        preset = get_preset().model_copy(update={"name": "qwen"})
         PresetStore(tmp_path / ".dstack" / "presets").save(preset)
 
         assert run_dstack_cli(["preset", "get", "qwen", "--json"], home_dir=tmp_path) == 0
diff --git a/src/tests/_internal/cli/models/test_configurations.py b/src/tests/_internal/cli/models/test_configurations.py
index 81b17146d3..ebe39e2412 100644
--- a/src/tests/_internal/cli/models/test_configurations.py
+++ b/src/tests/_internal/cli/models/test_configurations.py
@@ -12,12 +12,12 @@
 
 class TestPresetConfiguration:
     def test_schema_documents_supported_input(self):
-        assert all(
-            field.field_info.description for field in PresetConfiguration.__fields__.values()
-        )
-        assert all(field.field_info.description for field in PresetModelBase.__fields__.values())
-        assert all(field.field_info.description for field in PresetModelRepo.__fields__.values())
-        assert {"type": "string"} in PresetConfiguration.schema()["properties"]["model"]["anyOf"]
+        assert all(field.description for field in PresetConfiguration.model_fields.values())
+        assert all(field.description for field in PresetModelBase.model_fields.values())
+        assert all(field.description for field in PresetModelRepo.model_fields.values())
+        assert {"type": "string"} in PresetConfiguration.model_json_schema()["properties"][
+            "model"
+        ]["anyOf"]
 
     def test_parses_string_as_exact_repo(self):
         configuration = PresetConfiguration(model="Qwen/Qwen3.5-27B")
@@ -67,7 +67,7 @@ def test_parses_top_level_repo_shorthand(self):
     def test_shorthand_round_trips_through_dict(self):
         configuration = PresetConfiguration(base="Qwen/Qwen3.5-27B")
 
-        round_tripped = PresetConfiguration.parse_obj(configuration.dict())
+        round_tripped = PresetConfiguration.model_validate(configuration.model_dump())
 
         assert round_tripped.model == configuration.model
 
diff --git a/src/tests/_internal/cli/models/test_presets.py b/src/tests/_internal/cli/models/test_presets.py
index d71e7eba8d..856f13bb91 100644
--- a/src/tests/_internal/cli/models/test_presets.py
+++ b/src/tests/_internal/cli/models/test_presets.py
@@ -16,19 +16,19 @@
 class TestPresetBenchmark:
     def test_agent_schema_matches_benchmark_model(self):
         schema = AGENT_FINAL_REPORT_JSON_SCHEMA["properties"]["benchmark"]
-        assert set(schema["properties"]) == set(PresetBenchmark.__fields__) - {
+        assert set(schema["properties"]) == set(PresetBenchmark.model_fields) - {
             "target",
             "client",
         }
         assert set(schema["required"]) == set(schema["properties"])
         workload_schema = schema["properties"]["workload"]
         metrics_schema = schema["properties"]["metrics"]
-        assert set(workload_schema["properties"]) == set(PresetBenchmarkWorkload.__fields__)
+        assert set(workload_schema["properties"]) == set(PresetBenchmarkWorkload.model_fields)
         assert set(workload_schema["required"]) == set(workload_schema["properties"])
-        assert set(metrics_schema["properties"]) == set(PresetBenchmarkMetrics.__fields__)
+        assert set(metrics_schema["properties"]) == set(PresetBenchmarkMetrics.model_fields)
         assert set(metrics_schema["required"]) == set(metrics_schema["properties"])
         assert set(metrics_schema["properties"]["ttft_ms"]["properties"]) == set(
-            PresetBenchmarkLatency.__fields__
+            PresetBenchmarkLatency.model_fields
         )
 
     @pytest.mark.parametrize(
@@ -39,14 +39,16 @@ def test_agent_schema_matches_benchmark_model(self):
         ],
     )
     def test_rejects_inconsistent_successful_metrics(self, field, value, error):
-        data = get_preset_benchmark().dict()
+        data = get_preset_benchmark().model_dump()
         data["metrics"][field] = value
         with pytest.raises(ValidationError, match=error):
-            PresetBenchmark.parse_obj(data)
+            PresetBenchmark.model_validate(data)
 
     def test_rejects_tool_specific_metrics(self):
-        data = get_preset_benchmark().dict()
+        data = get_preset_benchmark().model_dump()
         data["metrics"]["tool_specific"] = 1
 
-        with pytest.raises(ValidationError, match="extra fields not permitted"):
-            PresetBenchmark.parse_obj(data)
+        # No `match=`: the wording is pydantic's, and v2 rewords it ("Extra inputs are not
+        # permitted" rather than "extra fields not permitted"). What matters is the rejection.
+        with pytest.raises(ValidationError):
+            PresetBenchmark.model_validate(data)
diff --git a/src/tests/_internal/cli/preset_factories.py b/src/tests/_internal/cli/preset_factories.py
index 3104447d7a..6d5f6cad1e 100644
--- a/src/tests/_internal/cli/preset_factories.py
+++ b/src/tests/_internal/cli/preset_factories.py
@@ -41,7 +41,7 @@ def get_preset_benchmark(*, verified: bool = True) -> PresetBenchmark:
     )
     if not verified:
         return benchmark
-    return benchmark.copy(
+    return benchmark.model_copy(
         update={
             "target": PresetBenchmarkTarget(type="server-proxy"),
             "client": PresetBenchmarkClient(type="local"),
@@ -54,7 +54,7 @@ def get_preset(
     preset_id: str = "8f3a12c4",
     context_length: int = 32768,
 ) -> Preset:
-    resources = ResourcesSpec.parse_obj(
+    resources = ResourcesSpec.model_validate(
         {
             "cpu": "16",
             "memory": "64GB",
@@ -68,7 +68,7 @@ def get_preset(
         model="community/Qwen3.5-27B-GPTQ-Int4",
         context_length=context_length,
         created_at=datetime(2026, 1, 2, 3, 4, tzinfo=timezone.utc),
-        service=ServiceConfiguration.parse_obj(
+        service=ServiceConfiguration.model_validate(
             {
                 "image": "vllm/vllm-openai:v0.11.0",
                 "commands": ["vllm serve community/Qwen3.5-27B-GPTQ-Int4"],
@@ -88,7 +88,7 @@ def get_preset(
 
 
 def get_running_service_run() -> Run:
-    service = ServiceConfiguration.parse_obj(
+    service = ServiceConfiguration.model_validate(
         {
             "name": "qwen-build-2",
             "image": "vllm/vllm-openai:v0.11.0",
@@ -125,7 +125,7 @@ def get_running_service_run() -> Run:
             )
         ],
     )
-    return Run.construct(
+    return Run.model_construct(
         id=uuid4(),
         project_name="main",
         status=RunStatus.RUNNING,
diff --git a/src/tests/_internal/cli/services/configurators/test_fleet.py b/src/tests/_internal/cli/services/configurators/test_fleet.py
index f1d0bfe222..80bc215e98 100644
--- a/src/tests/_internal/cli/services/configurators/test_fleet.py
+++ b/src/tests/_internal/cli/services/configurators/test_fleet.py
@@ -29,7 +29,7 @@
 
 
 def create_conf() -> FleetConfiguration:
-    return FleetConfiguration.parse_obj({"ssh_config": {"hosts": ["1.2.3.4"]}})
+    return FleetConfiguration.model_validate({"ssh_config": {"hosts": ["1.2.3.4"]}})
 
 
 def apply_args(
@@ -38,7 +38,7 @@ def apply_args(
     parser = argparse.ArgumentParser()
     configurator = FleetConfigurator(Mock())
     configurator.register_args(parser)
-    conf = conf.copy(deep=True)
+    conf = conf.model_copy(deep=True)
     configurator_args = parser.parse_args(args)
     configurator.apply_args(conf, configurator_args)
     return conf, configurator_args
@@ -71,7 +71,7 @@ def get_ssh_fleet_spec(
     if hosts is None:
         hosts = ["10.0.0.100"]
     return FleetSpec(
-        configuration=FleetConfiguration.parse_obj(
+        configuration=FleetConfiguration.model_validate(
             {
                 "name": name,
                 "ssh_config": {"hosts": hosts},
@@ -130,23 +130,23 @@ class TestFleetConfigurator:
     def test_env(self):
         conf = create_conf()
         modified, args = apply_args(conf, ["-e", "A=1", "--env", "B=2"])
-        conf.env = Env.parse_obj({"A": "1", "B": "2"})
-        assert modified.dict() == conf.dict()
+        conf.env = Env.model_validate({"A": "1", "B": "2"})
+        assert modified.model_dump() == conf.model_dump()
 
     def test_env_override(self):
         conf = create_conf()
-        conf.env = Env.parse_obj({"A": "0"})
+        conf.env = Env.model_validate({"A": "0"})
         modified, args = apply_args(conf, ["-e", "A=1", "--env", "B=2"])
-        conf.env = Env.parse_obj({"A": "1", "B": "2"})
-        assert modified.dict() == conf.dict()
+        conf.env = Env.model_validate({"A": "1", "B": "2"})
+        assert modified.model_dump() == conf.model_dump()
 
     def test_env_value_from_environ(self, monkeypatch: pytest.MonkeyPatch):
         monkeypatch.setenv("FROM_ENV", "2")
         conf = create_conf()
-        conf.env = Env.parse_obj({"FROM_CONF": "1"})
+        conf.env = Env.model_validate({"FROM_CONF": "1"})
         modified, args = apply_args(conf, ["--env", "FROM_ENV"])
-        conf.env = Env.parse_obj({"FROM_CONF": "1", "FROM_ENV": "2"})
-        assert modified.dict() == conf.dict()
+        conf.env = Env.model_validate({"FROM_CONF": "1", "FROM_ENV": "2"})
+        assert modified.model_dump() == conf.model_dump()
 
     def test_env_value_from_environ_not_set(self, monkeypatch: pytest.MonkeyPatch):
         monkeypatch.delenv("FROM_ENV", raising=False)
@@ -199,7 +199,7 @@ def test_prints_no_diff_message(self, monkeypatch: pytest.MonkeyPatch):
         spec = get_cloud_fleet_spec()
         plan = create_fleet_plan(
             current_spec=spec,
-            spec=spec.copy(deep=True),
+            spec=spec.model_copy(deep=True),
             action=ApplyAction.UPDATE,
         )
 
@@ -244,4 +244,4 @@ def test_renders_ssh_hosts_change(self):
     def test_no_diff(self):
         spec = get_cloud_fleet_spec()
 
-        assert _render_fleet_spec_diff(spec, spec.copy(deep=True)) is None
+        assert _render_fleet_spec_diff(spec, spec.model_copy(deep=True)) is None
diff --git a/src/tests/_internal/cli/services/configurators/test_profile.py b/src/tests/_internal/cli/services/configurators/test_profile.py
index d3c363c0f2..796b459b46 100644
--- a/src/tests/_internal/cli/services/configurators/test_profile.py
+++ b/src/tests/_internal/cli/services/configurators/test_profile.py
@@ -1,10 +1,13 @@
 import argparse
 from typing import List, Tuple
 
+import pytest
+
 from dstack._internal.cli.services.profile import (
     apply_profile_args,
     register_profile_args,
 )
+from dstack._internal.core.models.backends.base import BackendType
 from dstack._internal.core.models.profiles import Profile, ProfileRetry, SpotPolicy
 
 
@@ -12,67 +15,74 @@ class TestProfileArgs:
     def test_empty(self):
         profile = Profile(name="test")
         modified, _ = apply_args(profile, [])
-        assert profile.dict() == modified.dict()
+        assert profile.model_dump() == modified.model_dump()
 
     def test_profile_name(self):
         profile = Profile(name="test")
         modified, args = apply_args(profile, ["--profile", "test2"])
-        assert profile.dict() == modified.dict()
+        assert profile.model_dump() == modified.model_dump()
         assert args.profile == "test2"
 
     def test_max_price(self):
         profile = Profile(name="test")
         modified, _ = apply_args(profile, ["--max-price", "0.5"])
         profile.max_price = 0.5
-        assert profile.dict() == modified.dict()
+        assert profile.model_dump() == modified.model_dump()
 
     def test_max_duration(self):
         profile = Profile(name="test")
         modified, _ = apply_args(profile, ["--max-duration", "1h"])
         profile.max_duration = 3600
-        assert profile.dict() == modified.dict()
+        assert profile.model_dump() == modified.model_dump()
 
     def test_backends(self):
         profile = Profile(name="test")
         modified, _ = apply_args(profile, ["-b", "gcp", "--backend", "aws"])
-        profile.backends = ["gcp", "aws"]
-        assert profile.dict() == modified.dict()
+        # `BackendType`, not `str`: the args are assigned onto the model without validation, so
+        # anything but the declared type makes the serializer warn.
+        profile.backends = [BackendType.GCP, BackendType.AWS]
+        assert profile.model_dump() == modified.model_dump()
+
+    def test_backends_rejects_unknown_name(self, capsys: pytest.CaptureFixture):
+        with pytest.raises(SystemExit):
+            apply_args(Profile(name="test"), ["-b", "quantum-cloud-9000"])
+        assert "invalid BackendType value: 'quantum-cloud-9000'" in capsys.readouterr().err
 
     def test_spot_policy_spot(self):
         profile = Profile(name="test")
         modified, _ = apply_args(profile, ["--spot"])
         profile.spot_policy = SpotPolicy.SPOT
-        assert profile.dict() == modified.dict()
+        assert profile.model_dump() == modified.model_dump()
 
     def test_spot_policy_on_demand(self):
         profile = Profile(name="test")
         modified, _ = apply_args(profile, ["--on-demand"])
         profile.spot_policy = SpotPolicy.ONDEMAND
-        assert profile.dict() == modified.dict()
+        assert profile.model_dump() == modified.model_dump()
 
     def test_retry(self):
         profile = Profile(name="test", retry=None)
         modified, _ = apply_args(profile, ["--retry"])
         profile.retry = True
-        assert profile.dict() == modified.dict()
+        assert profile.model_dump() == modified.model_dump()
 
     def test_no_retry(self):
         profile = Profile(name="test", retry=None)
         modified, _ = apply_args(profile, ["--no-retry"])
         profile.retry = False
-        assert profile.dict() == modified.dict()
+        assert profile.model_dump() == modified.model_dump()
 
     def test_retry_duration(self):
         profile = Profile(name="test")
         modified, _ = apply_args(profile, ["--retry-duration", "1h"])
         profile.retry = ProfileRetry(on_events=None, duration="1h")
-        assert profile.dict() == modified.dict()
+        assert profile.model_dump() == modified.model_dump()
 
 
 def apply_args(profile: Profile, args: List[str]) -> Tuple[Profile, argparse.Namespace]:
     parser = argparse.ArgumentParser()
     register_profile_args(parser)
-    profile = profile.copy(deep=True)  # to avoid modifying the original profile
+    profile = profile.model_copy(deep=True)  # to avoid modifying the original profile
     parsed_args = parser.parse_args(args)
     apply_profile_args(parsed_args, profile)
     return profile, parsed_args
diff --git a/src/tests/_internal/cli/services/configurators/test_run.py b/src/tests/_internal/cli/services/configurators/test_run.py
index 6bc22b9703..c0d468feaf 100644
--- a/src/tests/_internal/cli/services/configurators/test_run.py
+++ b/src/tests/_internal/cli/services/configurators/test_run.py
@@ -38,7 +38,7 @@ def apply_args(
         configurator_class = get_run_configurator_class(conf.type)
         configurator = configurator_class(Mock())
         configurator.register_args(parser)
-        conf = conf.copy(deep=True)  # to avoid modifying the original configuration
+        conf = conf.model_copy(deep=True)  # to avoid modifying the original configuration
         parsed_args = parser.parse_args(args)
         configurator.apply_args(conf, parsed_args)
         return conf, parsed_args
@@ -46,8 +46,8 @@ def apply_args(
     def test_env(self):
         conf = TaskConfiguration(commands=["whoami"])
         modified, args = self.apply_args(conf, ["-e", "A=1", "--env", "B=2"])
-        conf.env = Env.parse_obj({"A": "1", "B": "2"})
-        assert modified.dict() == conf.dict()
+        conf.env = Env.model_validate({"A": "1", "B": "2"})
+        assert modified.model_dump() == conf.model_dump()
 
     def test_ports(self):
         conf = TaskConfiguration(commands=["whoami"])
@@ -56,7 +56,7 @@ def test_ports(self):
             PortMapping(local_port=80, container_port=80),
             PortMapping(local_port=8080, container_port=8080),
         ]
-        assert modified.dict() == conf.dict()
+        assert modified.model_dump() == conf.model_dump()
 
     def test_container_ports_conflict(self):
         conf = TaskConfiguration(commands=["whoami"])
@@ -64,10 +64,10 @@ def test_container_ports_conflict(self):
             self.apply_args(conf, ["-p", "8000:80", "--port", "8001:80"])
 
     def test_env_override(self):
-        conf = TaskConfiguration(commands=["whoami"], env=Env.parse_obj({"A": "0"}))
+        conf = TaskConfiguration(commands=["whoami"], env=Env.model_validate({"A": "0"}))
         modified, args = self.apply_args(conf, ["-e", "A=1", "--env", "B=2"])
-        conf.env = Env.parse_obj({"A": "1", "B": "2"})
-        assert modified.dict() == conf.dict()
+        conf.env = Env.model_validate({"A": "1", "B": "2"})
+        assert modified.model_dump() == conf.model_dump()
 
     def test_ports_override(self):
         conf = TaskConfiguration(commands=["whoami"], ports=["80"])
@@ -76,7 +76,7 @@ def test_ports_override(self):
             PortMapping(local_port=8000, container_port=80),
             PortMapping(local_port=8001, container_port=8000),
         ]
-        assert modified.dict() == conf.dict()
+        assert modified.model_dump() == conf.model_dump()
 
     def test_local_ports_conflict(self):
         conf = TaskConfiguration(commands=["whoami"], ports=["3000"])
@@ -87,7 +87,7 @@ def test_any_port(self):
         conf = TaskConfiguration(commands=["whoami"], ports=["8000"])
         modified, args = self.apply_args(conf, ["-p", "*:8000"])
         conf.ports = [PortMapping(local_port=None, container_port=8000)]
-        assert modified.dict() == conf.dict()
+        assert modified.model_dump() == conf.model_dump()
 
     def test_interpolates_env(self):
         conf = TaskConfiguration(
@@ -96,7 +96,7 @@ def test_interpolates_env(self):
                 username="${{ env.REGISTRY_USERNAME }}",
                 password="${{ env.REGISTRY_PASSWORD }}",
             ),
-            env=Env.parse_obj(
+            env=Env.model_validate(
                 {
                     "REGISTRY_USERNAME": "test_user",
                     "REGISTRY_PASSWORD": "test_password",
@@ -129,7 +129,7 @@ def prepare_conf(
             }
         if docker is not None:
             conf_dict["docker"] = docker
-        return BaseRunConfiguration.parse_obj(conf_dict)
+        return BaseRunConfiguration.model_validate(conf_dict)
 
     def validate(self, conf: BaseRunConfiguration) -> None:
         BaseRunConfigurator(api_client=Mock()).validate_gpu_vendor_and_image(conf)
@@ -302,7 +302,7 @@ def prepare_conf(
             conf_dict["image"] = image
         if gpu_spec is not None:
             conf_dict["resources"]["gpu"] = gpu_spec
-        return BaseRunConfiguration.parse_obj(conf_dict)
+        return BaseRunConfiguration.model_validate(conf_dict)
 
     def validate(self, conf: BaseRunConfiguration) -> None:
         # validate_gpu_vendor_and_image sets GPU vendor if not set
diff --git a/src/tests/_internal/cli/services/presets/test_agent.py b/src/tests/_internal/cli/services/presets/test_agent.py
index 9e96c44128..e3b17396d0 100644
--- a/src/tests/_internal/cli/services/presets/test_agent.py
+++ b/src/tests/_internal/cli/services/presets/test_agent.py
@@ -780,7 +780,7 @@ def test_loads_interrupted_session(self, tmp_path, monkeypatch):
                 "status": "interrupted",
                 "claude_session_id": "sid-1",
                 "debug": True,
-                "created_at": "2026-07-20T10:00:00+00:00",
+                "created_at": "2026-07-20T10:00:00Z",
             },
         )
 
diff --git a/src/tests/_internal/cli/services/presets/test_apply.py b/src/tests/_internal/cli/services/presets/test_apply.py
index 3eae3cf69d..919d2c02f9 100644
--- a/src/tests/_internal/cli/services/presets/test_apply.py
+++ b/src/tests/_internal/cli/services/presets/test_apply.py
@@ -51,7 +51,7 @@ def test_exact_request_matches_repo_and_client_facing_name(self):
         _validate_preset_matches(matching, configuration=configuration)
         with pytest.raises(CLIError, match="does not serve repo"):
             _validate_preset_matches(
-                matching.copy(update={"model": "other/repo"}),
+                matching.model_copy(update={"model": "other/repo"}),
                 configuration=configuration,
             )
 
diff --git a/src/tests/_internal/cli/services/presets/test_create.py b/src/tests/_internal/cli/services/presets/test_create.py
index e87e589f58..59f11968a2 100644
--- a/src/tests/_internal/cli/services/presets/test_create.py
+++ b/src/tests/_internal/cli/services/presets/test_create.py
@@ -315,7 +315,9 @@ async def run_agent(**kwargs):
             assert kwargs["agent_session"] is agent_session
             assert (session_path / "prompt.md").is_file()
             return PresetAgentProcessOutput(
-                report_data=json.loads(get_successful_preset_report(creation_context.run).json())
+                report_data=json.loads(
+                    get_successful_preset_report(creation_context.run).model_dump_json()
+                )
             )
 
         monkeypatch.setattr(
@@ -574,7 +576,9 @@ async def test_resume_uses_saved_claude_session(self, creation_context, monkeypa
         async def run_agent(**kwargs):
             captured.update(kwargs)
             return PresetAgentProcessOutput(
-                report_data=json.loads(get_successful_preset_report(creation_context.run).json())
+                report_data=json.loads(
+                    get_successful_preset_report(creation_context.run).model_dump_json()
+                )
             )
 
         monkeypatch.setattr(
@@ -608,7 +612,9 @@ async def test_pins_user_prompt_on_create(self, creation_context, monkeypatch, t
         async def run_agent(**kwargs):
             captured.update(kwargs)
             return PresetAgentProcessOutput(
-                report_data=json.loads(get_successful_preset_report(creation_context.run).json())
+                report_data=json.loads(
+                    get_successful_preset_report(creation_context.run).model_dump_json()
+                )
             )
 
         monkeypatch.setattr(
@@ -649,7 +655,9 @@ async def test_resume_keeps_the_pinned_user_prompt(
         async def run_agent(**kwargs):
             captured.update(kwargs)
             return PresetAgentProcessOutput(
-                report_data=json.loads(get_successful_preset_report(creation_context.run).json())
+                report_data=json.loads(
+                    get_successful_preset_report(creation_context.run).model_dump_json()
+                )
             )
 
         monkeypatch.setattr(
@@ -756,7 +764,9 @@ def test_finalizes_a_detached_session(self, creation_context, monkeypatch, tmp_p
 
         async def fake_attach(**kwargs):
             return PresetAgentProcessOutput(
-                report_data=json.loads(get_successful_preset_report(creation_context.run).json())
+                report_data=json.loads(
+                    get_successful_preset_report(creation_context.run).model_dump_json()
+                )
             )
 
         monkeypatch.setattr(
diff --git a/src/tests/_internal/cli/services/presets/test_output.py b/src/tests/_internal/cli/services/presets/test_output.py
index 7e76dbe063..49cca63378 100644
--- a/src/tests/_internal/cli/services/presets/test_output.py
+++ b/src/tests/_internal/cli/services/presets/test_output.py
@@ -110,19 +110,21 @@ def test_sorts_presets_and_sessions_newest_first(self, monkeypatch):
         buffer = StringIO()
         monkeypatch.setattr(output_module, "console", plain_console(buffer, width=200))
         old = get_preset()
-        new = old.copy(update={"id": "11aa22bb", "created_at": old.created_at + timedelta(days=2)})
+        new = old.model_copy(
+            update={"id": "11aa22bb", "created_at": old.created_at + timedelta(days=2)}
+        )
         sessions = [
             {
                 "id": "aaaaaaaa",
                 "status": "interrupted",
                 "model": old.base,
-                "created_at": "2026-07-01T00:00:00+00:00",
+                "created_at": "2026-07-01T00:00:00Z",
             },
             {
                 "id": "bbbbbbbb",
                 "status": "interrupted",
                 "model": old.base,
-                "created_at": "2026-07-02T00:00:00+00:00",
+                "created_at": "2026-07-02T00:00:00Z",
             },
         ]
 
diff --git a/src/tests/_internal/cli/services/presets/test_store.py b/src/tests/_internal/cli/services/presets/test_store.py
index a5b66cfe74..dcf6f05e53 100644
--- a/src/tests/_internal/cli/services/presets/test_store.py
+++ b/src/tests/_internal/cli/services/presets/test_store.py
@@ -38,7 +38,7 @@ def test_saving_same_id_overwrites_existing_preset(self, tmp_path: Path):
         preset = get_preset()
         store.save(preset)
 
-        updated = preset.copy(update={"base": "Qwen/Another-Model", "context_length": 16384})
+        updated = preset.model_copy(update={"base": "Qwen/Another-Model", "context_length": 16384})
         store.save(updated)
 
         assert store.get(preset.id) == updated
@@ -67,7 +67,7 @@ def test_migrates_legacy_layout_and_archives_on_delete(self, tmp_path: Path):
 
     def test_skips_invalid_preset_on_list_but_keeps_it_deletable(self, tmp_path: Path, capsys):
         store = PresetStore(tmp_path / "presets")
-        valid = get_preset().copy(update={"id": "01234567"})
+        valid = get_preset().model_copy(update={"id": "01234567"})
         store.save(valid)
         path = store.save(get_preset())
         data = yaml.safe_load(path.read_text())
@@ -168,9 +168,9 @@ def test_rejects_missing_and_empty_prompt_files(self, tmp_path: Path):
 class TestPresetNames:
     def test_finds_and_detaches_names(self, tmp_path: Path):
         store = PresetStore(tmp_path / "presets")
-        named = get_preset().copy(update={"name": "qwen"})
+        named = get_preset().model_copy(update={"name": "qwen"})
         store.save(named)
-        store.save(get_preset().copy(update={"id": "01234567"}))
+        store.save(get_preset().model_copy(update={"id": "01234567"}))
 
         assert store.find_by_name("qwen").id == named.id
         assert store.find_by_name("other") is None
diff --git a/src/tests/_internal/cli/services/presets/test_verify.py b/src/tests/_internal/cli/services/presets/test_verify.py
index 793bcda71c..fda1790b85 100644
--- a/src/tests/_internal/cli/services/presets/test_verify.py
+++ b/src/tests/_internal/cli/services/presets/test_verify.py
@@ -30,11 +30,11 @@
 class TestBuildVerifiedPreset:
     def test_successful_report_requires_benchmark(self):
         run = get_running_service_run()
-        data = get_successful_preset_report(run).dict()
+        data = get_successful_preset_report(run).model_dump()
         data.pop("benchmark")
 
         with pytest.raises(ValidationError, match="benchmark"):
-            AgentFinalReport.parse_obj(data)
+            AgentFinalReport.model_validate(data)
 
     def test_builds_portable_self_contained_preset(self):
         run = get_running_service_run()
@@ -62,7 +62,7 @@ def test_builds_portable_self_contained_preset(self):
         assert preset.created_at == created_at
         assert preset.service.name is None
         assert preset.service.gateway is None
-        assert all(getattr(preset.service, field) is None for field in ProfileParams.__fields__)
+        assert all(getattr(preset.service, field) is None for field in ProfileParams.model_fields)
         assert isinstance(preset.service.env["LICENSE"], EnvSentinel)
         assert preset.service.env["TOKENIZERS_PARALLELISM"] == "false"
         assert preset.service.resources.gpu.vendor.value == "nvidia"
@@ -73,7 +73,7 @@ def test_builds_portable_self_contained_preset(self):
 
     def test_rejects_variant_for_exact_model_request(self):
         run = get_running_service_run()
-        report = get_successful_preset_report(run).copy(update={"model": "other/model"})
+        report = get_successful_preset_report(run).model_copy(update={"model": "other/model"})
 
         with pytest.raises(CLIError, match="changed an exact model request"):
             build_verified_preset(
@@ -99,7 +99,7 @@ def _load(self, tmp_path, report_data, redacted_values):
 
     def test_redacts_known_secret_in_benchmark_command_instead_of_failing(self, tmp_path):
         run = get_running_service_run()
-        data = get_successful_preset_report(run).dict()
+        data = get_successful_preset_report(run).model_dump()
         data["run_id"] = str(data["run_id"])
         data["benchmark"]["command"] = (
             "python bench.py --header 'Authorization: Bearer sk-live-0123456789abcdef'"
@@ -113,7 +113,7 @@ def test_redacts_known_secret_in_benchmark_command_instead_of_failing(self, tmp_
 
     def test_still_rejects_unknown_bearer_token(self, tmp_path):
         run = get_running_service_run()
-        data = get_successful_preset_report(run).dict()
+        data = get_successful_preset_report(run).model_dump()
         data["run_id"] = str(data["run_id"])
         data["benchmark"]["command"] = (
             "curl -H 'Authorization: Bearer sk-unknown-9876543210fedcba'"
@@ -126,7 +126,7 @@ def test_allows_bearer_prose_without_credential(self, tmp_path):
         # Regression: "(auth via DSTACK_TOKEN bearer header from env)" failed
         # two live sessions — the word after "bearer" is prose, not a token.
         run = get_running_service_run()
-        data = get_successful_preset_report(run).dict()
+        data = get_successful_preset_report(run).model_dump()
         data["run_id"] = str(data["run_id"])
         data["benchmark"]["command"] = (
             "./benchenv/bin/python bench_service.py --base $DSTACK_SERVER_URL/x"
diff --git a/src/tests/_internal/cli/utils/conftest.py b/src/tests/_internal/cli/utils/conftest.py
index 0f87c88bd0..a374bf9d3a 100644
--- a/src/tests/_internal/cli/utils/conftest.py
+++ b/src/tests/_internal/cli/utils/conftest.py
@@ -7,7 +7,9 @@
 
 @pytest.fixture
 def image_config_mock(monkeypatch: pytest.MonkeyPatch) -> ImageConfig:
-    image_config = ImageConfig.parse_obj({"User": None, "Entrypoint": None, "Cmd": ["/bin/bash"]})
+    image_config = ImageConfig.model_validate(
+        {"User": None, "Entrypoint": None, "Cmd": ["/bin/bash"]}
+    )
     monkeypatch.setattr(
         "dstack._internal.server.services.jobs.configurators.base._get_image_config",
         Mock(return_value=image_config),
diff --git a/src/tests/_internal/core/backends/jarvislabs/test_compute.py b/src/tests/_internal/core/backends/jarvislabs/test_compute.py
index e9b055cd21..10ffae269f 100644
--- a/src/tests/_internal/core/backends/jarvislabs/test_compute.py
+++ b/src/tests/_internal/core/backends/jarvislabs/test_compute.py
@@ -384,7 +384,7 @@ def test_terminate_instance_deletes_created_ssh_keys():
     compute = _compute()
     backend_data = JarvisLabsInstanceBackendData(
         ssh_key_ids=["ssh-key-id-1", "ssh-key-id-2"]
-    ).json()
+    ).model_dump_json()
 
     compute.terminate_instance("123", "india-noida-01", backend_data)
 
diff --git a/src/tests/_internal/core/backends/slurm/test_configurator.py b/src/tests/_internal/core/backends/slurm/test_configurator.py
index bf576a80e6..c3b2895f01 100644
--- a/src/tests/_internal/core/backends/slurm/test_configurator.py
+++ b/src/tests/_internal/core/backends/slurm/test_configurator.py
@@ -36,8 +36,8 @@ def test_without_creds_strips_sensitive_fields(self):
         cluster = config.clusters[0]
         # The credentialless cluster config exposes only non-sensitive fields; connection
         # details and the private key must not be present at all.
-        assert set(cluster.__fields__) == {"name", "gpu_partitions", "cpu_partitions"}
-        rendered = config.json()
+        assert set(type(cluster).model_fields) == {"name", "gpu_partitions", "cpu_partitions"}
+        rendered = config.model_dump_json()
         for secret in (PRIVATE_KEY, HOSTNAME, str(PORT), USER):
             assert secret not in rendered
 
diff --git a/src/tests/_internal/core/backends/verda/test_compute.py b/src/tests/_internal/core/backends/verda/test_compute.py
index b8ae8f494d..d7ddad065c 100644
--- a/src/tests/_internal/core/backends/verda/test_compute.py
+++ b/src/tests/_internal/core/backends/verda/test_compute.py
@@ -354,7 +354,7 @@ def test_terminate_instance_deletes_startup_script(self):
         backend_data = VerdaInstanceBackendData(
             startup_script_id="script-id",
             ssh_key_ids=["ssh-key-id-1", "ssh-key-id-2"],
-        ).json()
+        ).model_dump_json()
 
         compute.terminate_instance("instance-id", "FIN-01", backend_data)
 
@@ -369,7 +369,7 @@ def test_terminate_instance_still_deletes_script_when_instance_is_missing(self):
         backend_data = VerdaInstanceBackendData(
             startup_script_id="script-id",
             ssh_key_ids=["ssh-key-id-1"],
-        ).json()
+        ).model_dump_json()
 
         compute.terminate_instance("instance-id", "FIN-01", backend_data)
 
@@ -385,7 +385,7 @@ def test_terminate_instance_retries_on_script_delete_error(self):
         backend_data = VerdaInstanceBackendData(
             startup_script_id="script-id",
             ssh_key_ids=["ssh-key-id-1"],
-        ).json()
+        ).model_dump_json()
 
         with pytest.raises(APIException):
             compute.terminate_instance("instance-id", "FIN-01", backend_data)
@@ -399,7 +399,7 @@ def test_terminate_instance_retries_on_ssh_key_delete_error(self):
         backend_data = VerdaInstanceBackendData(
             startup_script_id="script-id",
             ssh_key_ids=["ssh-key-id-1"],
-        ).json()
+        ).model_dump_json()
 
         with pytest.raises(APIException):
             compute.terminate_instance("instance-id", "FIN-01", backend_data)
diff --git a/src/tests/_internal/core/models/test_configurations.py b/src/tests/_internal/core/models/test_configurations.py
index 04095777ef..98f6ac2b5b 100644
--- a/src/tests/_internal/core/models/test_configurations.py
+++ b/src/tests/_internal/core/models/test_configurations.py
@@ -124,10 +124,12 @@ def test_conf(replicas: Any, scaling: Optional[Any] = None):
                 },
             )
         ).replicas == Range(min=0, max=10)
-        with pytest.raises(
-            ConfigurationError,
-            match="When you set `replicas` to a range, ensure to specify `scaling`",
-        ):
+        # `metric: rpc` is a typo, so `scaling` itself fails to validate. The config stays
+        # rejected, but the message is no longer the `scaling`-is-missing one: pydantic v1 ran a
+        # bare `root_validator` even after a field had failed, handing it a partial `values` dict
+        # in which `scaling` was absent. A v2 `model_validator(mode="after")` does not run at all
+        # once a field is invalid, so what surfaces is the actual typo — the more precise error.
+        with pytest.raises(ConfigurationError, match="metric"):
             parse_run_configuration(
                 test_conf(
                     "0..10",
diff --git a/src/tests/_internal/core/models/test_duration.py b/src/tests/_internal/core/models/test_duration.py
new file mode 100644
index 0000000000..dc74d4255f
--- /dev/null
+++ b/src/tests/_internal/core/models/test_duration.py
@@ -0,0 +1,104 @@
+from typing import Any, Optional
+
+import pytest
+from pydantic import TypeAdapter, ValidationError
+
+from dstack._internal.core.models.duration import (
+    Duration,
+    OptionalIdleDuration,
+    OptionalOffableDuration,
+)
+
+
+class TestDuration:
+    @pytest.mark.parametrize(
+        ("raw", "expected"),
+        [
+            pytest.param(90, 90, id="int"),
+            pytest.param("90", 90, id="numeric-string"),
+            pytest.param(1.9, 1, id="float-truncates"),
+            pytest.param("30s", 30, id="seconds"),
+            pytest.param("5m", 300, id="minutes"),
+            pytest.param("2h", 7200, id="hours"),
+            pytest.param("2 h", 7200, id="space-before-unit"),
+            pytest.param("1d", 86400, id="days"),
+            pytest.param("1w", 604800, id="weeks"),
+        ],
+    )
+    def test_parses_seconds_and_shorthands(self, raw: Any, expected: int):
+        value = TypeAdapter(Duration).validate_python(raw)
+
+        assert value == expected
+        assert isinstance(value, Duration)
+
+    @pytest.mark.parametrize(
+        "raw",
+        [
+            pytest.param("bogus", id="not-a-duration"),
+            pytest.param("2y", id="unsupported-unit"),
+            pytest.param("h2", id="unit-before-amount"),
+            pytest.param("", id="empty"),
+        ],
+    )
+    def test_rejects_unparsable(self, raw: str):
+        with pytest.raises(ValidationError):
+            TypeAdapter(Duration).validate_python(raw)
+
+    def test_serializes_as_seconds(self):
+        adapter = TypeAdapter(Duration)
+
+        assert adapter.dump_python(Duration.parse("2h")) == 7200
+        assert adapter.dump_json(Duration.parse("2h")) == b"7200"
+
+
+class TestOptionalOffableDuration:
+    @pytest.mark.parametrize(
+        ("raw", "expected"),
+        [
+            pytest.param("2h", 7200, id="shorthand"),
+            pytest.param(7200, 7200, id="seconds"),
+            pytest.param("off", "off", id="off-string"),
+            pytest.param(False, "off", id="false-means-off"),
+            pytest.param(True, None, id="true-means-default"),
+            pytest.param(None, None, id="unspecified"),
+        ],
+    )
+    def test_normalizes(self, raw: Any, expected: Any):
+        assert TypeAdapter(OptionalOffableDuration).validate_python(raw) == expected
+
+    @pytest.mark.parametrize("raw", [-1, -300, "-1"])
+    def test_rejects_negative(self, raw: Any):
+        """Unlike `OptionalIdleDuration`, there is no negative sentinel here."""
+        with pytest.raises(ValidationError):
+            TypeAdapter(OptionalOffableDuration).validate_python(raw)
+
+
+class TestOptionalIdleDuration:
+    @pytest.mark.parametrize(
+        ("raw", "expected"),
+        [
+            pytest.param("5m", 300, id="shorthand"),
+            pytest.param(300, 300, id="seconds"),
+            pytest.param("off", -1, id="off-string"),
+            pytest.param(False, -1, id="false-means-off"),
+            pytest.param(True, None, id="true-means-default"),
+            pytest.param(None, None, id="unspecified"),
+        ],
+    )
+    def test_normalizes(self, raw: Any, expected: Optional[int]):
+        assert TypeAdapter(OptionalIdleDuration).validate_python(raw) == expected
+
+    @pytest.mark.parametrize(
+        "raw",
+        [
+            pytest.param(-1, id="minus-one"),
+            pytest.param(-300, id="other-negative"),
+            pytest.param("-1", id="minus-one-string"),
+        ],
+    )
+    def test_accepts_negative_for_backward_compatibility(self, raw: Any):
+        """
+        `-1` is how older clients and existing stored rows spell "off", so negatives have to keep
+        parsing rather than being rejected the way `OptionalOffableDuration` rejects them.
+        """
+        assert TypeAdapter(OptionalIdleDuration).validate_python(raw) < 0
diff --git a/src/tests/_internal/core/models/test_fleets.py b/src/tests/_internal/core/models/test_fleets.py
index a9214f7ece..83be0cabb5 100644
--- a/src/tests/_internal/core/models/test_fleets.py
+++ b/src/tests/_internal/core/models/test_fleets.py
@@ -99,7 +99,7 @@ def test_parses_nodes(self, input_nodes: Any, expected_nodes: FleetNodesSpec):
             "type": "fleet",
             "nodes": input_nodes,
         }
-        configuration = FleetConfiguration.parse_obj(configuration_input)
+        configuration = FleetConfiguration.model_validate(configuration_input)
         assert configuration.nodes == expected_nodes
 
     @pytest.mark.parametrize(
@@ -119,4 +119,4 @@ def test_rejects_nodes(self, input_nodes: Any):
             "nodes": input_nodes,
         }
         with pytest.raises(ValidationError):
-            FleetConfiguration.parse_obj(configuration_input)
+            FleetConfiguration.model_validate(configuration_input)
diff --git a/src/tests/_internal/core/models/test_instances.py b/src/tests/_internal/core/models/test_instances.py
index bedc708fa0..75c21b9e61 100644
--- a/src/tests/_internal/core/models/test_instances.py
+++ b/src/tests/_internal/core/models/test_instances.py
@@ -5,7 +5,7 @@
 
 class TestGpu:
     def test_no_vendor_nvidia(self):
-        gpu = Gpu.parse_obj(
+        gpu = Gpu.model_validate(
             {
                 "name": "T4",
                 "memory_mib": 16,
@@ -15,7 +15,7 @@ def test_no_vendor_nvidia(self):
         assert gpu.name == "T4"
 
     def test_no_vendor_tpu(self):
-        gpu = Gpu.parse_obj(
+        gpu = Gpu.model_validate(
             {
                 "name": "tpu-v3",
                 "memory_mib": 0,
@@ -25,7 +25,7 @@ def test_no_vendor_tpu(self):
         assert gpu.name == "v3"
 
     def test_vendor_cast_to_enum(self):
-        gpu = Gpu.parse_obj(
+        gpu = Gpu.model_validate(
             {
                 "vendor": "AMD",
                 "name": "MI300X",
diff --git a/src/tests/_internal/core/models/test_profiles.py b/src/tests/_internal/core/models/test_profiles.py
index 246435f6f4..fc789c8251 100644
--- a/src/tests/_internal/core/models/test_profiles.py
+++ b/src/tests/_internal/core/models/test_profiles.py
@@ -38,7 +38,7 @@ def test_empty_list_backend_options_is_valid(self):
 
 class TestProfileInstances:
     def test_string_is_parsed_as_instance_name_selector(self):
-        profile = Profile.parse_obj({"instances": ["my-fleet-1"]})
+        profile = Profile.model_validate({"instances": ["my-fleet-1"]})
 
         assert profile.instances == [InstanceNameSelector(name="my-fleet-1")]
 
@@ -58,12 +58,12 @@ def test_string_is_parsed_as_instance_name_selector(self):
         ],
     )
     def test_object_selectors_are_parsed(self, value, expected):
-        profile = Profile.parse_obj({"instances": [value]})
+        profile = Profile.model_validate({"instances": [value]})
 
         assert profile.instances == [expected]
 
     def test_parses_fleet_selector_object_notation(self):
-        profile = Profile.parse_obj(
+        profile = Profile.model_validate(
             {"instances": [{"fleet": {"project": "main", "name": "my-fleet"}, "instance": 0}]}
         )
 
@@ -89,22 +89,22 @@ def test_parses_fleet_selector_object_notation(self):
     )
     def test_invalid_selector_is_rejected(self, value):
         with pytest.raises(ValidationError):
-            Profile.parse_obj({"instances": [value]})
+            Profile.model_validate({"instances": [value]})
 
     def test_empty_instances_list_is_rejected(self):
         with pytest.raises(ValidationError):
-            Profile.parse_obj({"instances": []})
+            Profile.model_validate({"instances": []})
 
 
 class TestProfileInstancesCompatibilityExcludes:
     def test_excludes_unset_instances(self):
         profile = Profile()
 
-        assert "instances" not in profile.dict(exclude=get_profile_excludes(profile))
+        assert "instances" not in profile.model_dump(exclude=get_profile_excludes(profile))
 
     def test_preserves_configured_instances(self):
         profile = Profile(instances=[InstanceNameSelector(name="my-fleet-1")])
 
-        assert profile.dict(exclude=get_profile_excludes(profile))["instances"] == [
+        assert profile.model_dump(exclude=get_profile_excludes(profile))["instances"] == [
             {"name": "my-fleet-1"}
         ]
diff --git a/src/tests/_internal/core/models/test_resources.py b/src/tests/_internal/core/models/test_resources.py
index 5da32ec8f2..60227deef8 100644
--- a/src/tests/_internal/core/models/test_resources.py
+++ b/src/tests/_internal/core/models/test_resources.py
@@ -2,7 +2,7 @@
 
 import pytest
 from gpuhunt import AcceleratorVendor, CPUArchitecture
-from pydantic import ValidationError, parse_obj_as
+from pydantic import TypeAdapter, ValidationError
 
 from dstack._internal.core.models.resources import (
     DEFAULT_CPU_COUNT,
@@ -16,102 +16,107 @@
 
 class TestMemory:
     def test_mb(self):
-        assert parse_obj_as(Memory, "512MB") == 0.5
+        assert TypeAdapter(Memory).validate_python("512MB") == 0.5
 
     def test_gb(self):
-        assert parse_obj_as(Memory, "16 Gb") == 16.0
+        assert TypeAdapter(Memory).validate_python("16 Gb") == 16.0
 
     def test_tb(self):
-        assert parse_obj_as(Memory, "1 TB ") == 1024.0
+        assert TypeAdapter(Memory).validate_python("1 TB ") == 1024.0
 
     def test_float(self):
-        assert parse_obj_as(Memory, 1.5) == 1.5
+        assert TypeAdapter(Memory).validate_python(1.5) == 1.5
 
     def test_int(self):
-        assert parse_obj_as(Memory, 1) == 1.0
+        assert TypeAdapter(Memory).validate_python(1) == 1.0
 
     def test_invalid(self):
         with pytest.raises(ValidationError):
-            parse_obj_as(Memory, "1.5xb")
+            TypeAdapter(Memory).validate_python("1.5xb")
 
 
 class TestComputeCapability:
     def test_str(self):
-        assert parse_obj_as(ComputeCapability, "3.5") == (3, 5)
+        assert TypeAdapter(ComputeCapability).validate_python("3.5") == (3, 5)
 
     def test_float(self):
-        assert parse_obj_as(ComputeCapability, 8.0) == (8, 0)
+        assert TypeAdapter(ComputeCapability).validate_python(8.0) == (8, 0)
 
     def test_tuple(self):
-        assert parse_obj_as(ComputeCapability, (7, 5)) == (7, 5)
+        assert TypeAdapter(ComputeCapability).validate_python((7, 5)) == (7, 5)
 
     def test_invalid_len(self):
         with pytest.raises(ValidationError):
-            parse_obj_as(ComputeCapability, "3.5.1")
+            TypeAdapter(ComputeCapability).validate_python("3.5.1")
 
     def test_invalid_type(self):
         with pytest.raises(ValidationError):
-            parse_obj_as(ComputeCapability, "3.x")
+            TypeAdapter(ComputeCapability).validate_python("3.x")
 
 
 class TestIntRange:
     def test_int(self):
-        assert parse_obj_as(Range[int], 1).dict() == dict(min=1, max=1)
+        assert Range[int].model_validate(1).model_dump() == dict(min=1, max=1)
 
     def test_exact(self):
-        assert parse_obj_as(Range[int], "1").dict() == dict(min=1, max=1)
+        assert Range[int].model_validate("1").model_dump() == dict(min=1, max=1)
 
     def test_from(self):
-        assert parse_obj_as(Range[int], "1..").dict() == dict(min=1, max=None)
+        assert Range[int].model_validate("1..").model_dump() == dict(min=1, max=None)
 
     def test_to(self):
-        assert parse_obj_as(Range[int], "..1").dict() == dict(min=None, max=1)
+        assert Range[int].model_validate("..1").model_dump() == dict(min=None, max=1)
 
     def test_invalid_range(self):
         with pytest.raises(ValidationError):
-            parse_obj_as(Range[int], "..")
+            Range[int].model_validate("..")
 
     def test_range_typo(self):
         with pytest.raises(ValidationError):
-            parse_obj_as(Range[int], "1...3")
+            Range[int].model_validate("1...3")
 
     def test_dict(self):
-        assert parse_obj_as(Range[int], {"min": 1, "max": 3}).dict() == dict(min=1, max=3)
+        assert Range[int].model_validate({"min": 1, "max": 3}).model_dump() == dict(min=1, max=3)
 
     def test_unordered(self):
         with pytest.raises(ValidationError):
-            parse_obj_as(Range[int], "3..1")
+            Range[int].model_validate("3..1")
 
     def test__str__(self):
-        assert isinstance(str(parse_obj_as(Range[int], "1")), str)
+        assert isinstance(str(Range[int].model_validate("1")), str)
 
 
 class TestMemoryRange:
     def test_mb(self):
-        assert parse_obj_as(Range[Memory], "512MB").dict() == dict(min=0.5, max=0.5)
+        assert Range[Memory].model_validate("512MB").model_dump() == dict(min=0.5, max=0.5)
 
     def test_from(self):
-        assert parse_obj_as(Range[Memory], "512MB..").dict() == dict(min=0.5, max=None)
+        assert Range[Memory].model_validate("512MB..").model_dump() == dict(min=0.5, max=None)
 
     def test_to(self):
-        assert parse_obj_as(Range[Memory], "..1 TB").dict() == dict(min=None, max=1024.0)
+        assert Range[Memory].model_validate("..1 TB").model_dump() == dict(min=None, max=1024.0)
 
     def test_range(self):
-        assert parse_obj_as(Range[Memory], "512..1 TB").dict() == dict(min=512.0, max=1024.0)
+        assert Range[Memory].model_validate("512..1 TB").model_dump() == dict(
+            min=512.0, max=1024.0
+        )
 
     def test_invalid_range(self):
         with pytest.raises(ValidationError):
-            parse_obj_as(Range[Memory], "...")
+            Range[Memory].model_validate("...")
 
     def test_dict(self):
-        assert parse_obj_as(Range[Memory], {"min": "512MB", "max": "1TB"}).dict() == dict(
+        assert Range[Memory].model_validate({"min": "512MB", "max": "1TB"}).model_dump() == dict(
             min=0.5, max=1024.0
         )
 
 
 class TestCPU:
     def test_integer(self):
-        assert parse_obj_as(CPUSpec, 1).dict() == {"arch": None, "count": {"min": 1, "max": 1}}
+        assert CPUSpec.model_validate(1).model_dump() == {
+            "arch": None,
+            "count": {"min": 1, "max": 1},
+        }
 
     @pytest.mark.parametrize(
         ["value", "expected_arch", "expected_min", "expected_max"],
@@ -129,7 +134,7 @@ def test_valid_string(
         expected_min: Optional[int],
         expected_max: Optional[int],
     ):
-        assert parse_obj_as(CPUSpec, value).dict() == {
+        assert CPUSpec.model_validate(value).model_dump() == {
             "arch": expected_arch,
             "count": {"min": expected_min, "max": expected_max},
         }
@@ -145,34 +150,36 @@ def test_valid_string(
     )
     def test_invalid_string(self, value: str, error: str):
         with pytest.raises(ValidationError, match=error):
-            parse_obj_as(CPUSpec, value)
+            CPUSpec.model_validate(value)
 
     def test_range_object(self):
-        assert parse_obj_as(CPUSpec, Range[int](min=1, max=2)).dict() == {
+        assert CPUSpec.model_validate(Range[int](min=1, max=2)).model_dump() == {
             "arch": None,
             "count": {"min": 1, "max": 2},
         }
 
     def test_range_dict(self):
-        assert parse_obj_as(CPUSpec, {"min": 1, "max": 2}).dict() == {
+        assert CPUSpec.model_validate({"min": 1, "max": 2}).model_dump() == {
             "arch": None,
             "count": {"min": 1, "max": 2},
         }
 
     def test_valid_dict(self):
-        assert parse_obj_as(CPUSpec, {"arch": "ARM", "count": {"min": 1, "max": 2}}).dict() == {
+        assert CPUSpec.model_validate(
+            {"arch": "ARM", "count": {"min": 1, "max": 2}}
+        ).model_dump() == {
             "arch": CPUArchitecture.ARM,
             "count": {"min": 1, "max": 2},
         }
 
     def test_invalid_dict(self):
         with pytest.raises(ValidationError):
-            parse_obj_as(CPUSpec, {"arch": "x86", "min": 1, "max": 2})
+            CPUSpec.model_validate({"arch": "x86", "min": 1, "max": 2})
 
 
 class TestGPU:
     def test_count(self):
-        assert parse_obj_as(GPUSpec, "1") == parse_obj_as(GPUSpec, {"count": 1})
+        assert GPUSpec.model_validate("1") == GPUSpec.model_validate({"count": 1})
 
     @pytest.mark.parametrize(
         ["value", "expected"],
@@ -203,7 +210,7 @@ def test_count(self):
         ],
     )
     def test_vendor_in_string_form(self, value, expected):
-        assert parse_obj_as(GPUSpec, value) == parse_obj_as(GPUSpec, expected)
+        assert GPUSpec.model_validate(value) == GPUSpec.model_validate(expected)
 
     @pytest.mark.parametrize(
         ["value", "expected"],
@@ -218,44 +225,44 @@ def test_vendor_in_string_form(self, value, expected):
         ],
     )
     def test_vendor_in_object_form(self, value, expected):
-        assert parse_obj_as(GPUSpec, {"vendor": value}) == parse_obj_as(
-            GPUSpec, {"vendor": expected}
+        assert GPUSpec.model_validate({"vendor": value}) == GPUSpec.model_validate(
+            {"vendor": expected}
         )
 
     def test_name(self):
-        assert parse_obj_as(GPUSpec, "A100") == parse_obj_as(GPUSpec, {"name": ["A100"]})
+        assert GPUSpec.model_validate("A100") == GPUSpec.model_validate({"name": ["A100"]})
 
     def test_name_with_tpu_prefix(self):
-        spec = parse_obj_as(GPUSpec, "tpu-v3-2048")
+        spec = GPUSpec.model_validate("tpu-v3-2048")
         assert spec.name == ["v3-2048"]
 
     def test_memory(self):
-        assert parse_obj_as(GPUSpec, "16GB") == parse_obj_as(GPUSpec, {"memory": "16GB"})
+        assert GPUSpec.model_validate("16GB") == GPUSpec.model_validate({"memory": "16GB"})
 
     def test_names_count(self):
-        assert parse_obj_as(GPUSpec, "A10,A10G:2") == parse_obj_as(
-            GPUSpec, {"name": ["A10", "A10G"], "count": 2}
+        assert GPUSpec.model_validate("A10,A10G:2") == GPUSpec.model_validate(
+            {"name": ["A10", "A10G"], "count": 2}
         )
 
     def test_empty_name(self):
         with pytest.raises(ValidationError):
-            parse_obj_as(GPUSpec, "A100,:2")
+            GPUSpec.model_validate("A100,:2")
 
     def test_empty_token(self):
         with pytest.raises(ValidationError):
-            parse_obj_as(GPUSpec, "A100:")
+            GPUSpec.model_validate("A100:")
 
     def test_vendor_conflict(self):
         with pytest.raises(ValidationError, match=r"vendor conflict"):
-            parse_obj_as(GPUSpec, "Nvidia:A100:2:AMD")
+            GPUSpec.model_validate("Nvidia:A100:2:AMD")
 
     def test_count_conflict(self):
         with pytest.raises(ValidationError, match=r"count conflict"):
-            parse_obj_as(GPUSpec, "A100:2:3")
+            GPUSpec.model_validate("A100:2:3")
 
     def test_memory_range(self):
-        assert parse_obj_as(GPUSpec, "16GB..32") == parse_obj_as(
-            GPUSpec, {"memory": {"min": 16, "max": 32}}
+        assert GPUSpec.model_validate("16GB..32") == GPUSpec.model_validate(
+            {"memory": {"min": 16, "max": 32}}
         )
 
 
diff --git a/src/tests/_internal/core/models/test_runs.py b/src/tests/_internal/core/models/test_runs.py
index 973aaa9140..e0bb9fbcee 100644
--- a/src/tests/_internal/core/models/test_runs.py
+++ b/src/tests/_internal/core/models/test_runs.py
@@ -139,7 +139,7 @@ def test_dynamo_router_with_retry_at_profile_level_is_rejected(self):
             retry={"on_events": ["error"]},
         )
         with pytest.raises(ValidationError, match="Dynamo"):
-            RunSpec.parse_obj(spec)
+            RunSpec.model_validate(spec)
 
     def test_dynamo_router_with_retry_in_configuration_is_rejected(self):
         # retry can also be specified at configuration level; _merged_profile
@@ -150,12 +150,12 @@ def test_dynamo_router_with_retry_in_configuration_is_rejected(self):
             top_level_extras={"retry": {"on_events": ["error"]}},
         )
         with pytest.raises(ValidationError, match="Dynamo"):
-            RunSpec.parse_obj(spec)
+            RunSpec.model_validate(spec)
 
     def test_dynamo_router_without_retry_is_accepted(self):
         spec = _service_run_spec_dict(router_type="dynamo", retry=None)
         # Should not raise:
-        RunSpec.parse_obj(spec)
+        RunSpec.model_validate(spec)
 
     def test_sglang_router_with_retry_is_accepted(self):
         spec = _service_run_spec_dict(
@@ -163,11 +163,11 @@ def test_sglang_router_with_retry_is_accepted(self):
             retry={"on_events": ["error"]},
         )
         # SGLang services are unaffected by the validator.
-        RunSpec.parse_obj(spec)
+        RunSpec.model_validate(spec)
 
     def test_service_without_router_with_retry_is_accepted(self):
         spec = _service_run_spec_dict(router_type=None, retry={"on_events": ["error"]})
-        RunSpec.parse_obj(spec)
+        RunSpec.model_validate(spec)
 
     def test_non_service_run_with_retry_is_accepted(self):
         # Validator is service-only. A task or dev-environment with retry
@@ -184,4 +184,4 @@ def test_non_service_run_with_retry_is_accepted(self):
             "ssh_key_pub": "ssh-rsa AAAA...",
             "repo_data": {"repo_type": "virtual"},
         }
-        RunSpec.parse_obj(spec)
+        RunSpec.model_validate(spec)
diff --git a/src/tests/_internal/core/models/test_templates.py b/src/tests/_internal/core/models/test_templates.py
index e73decbb0c..96bf67cd5e 100644
--- a/src/tests/_internal/core/models/test_templates.py
+++ b/src/tests/_internal/core/models/test_templates.py
@@ -16,7 +16,7 @@
 class TestUITemplateParameter:
     def test_parses_name_parameter(self):
         data = {"type": "name"}
-        template = UITemplate.parse_obj(
+        template = UITemplate.model_validate(
             {
                 "type": "template",
                 "name": "t",
@@ -30,7 +30,7 @@ def test_parses_name_parameter(self):
 
     def test_parses_ide_parameter(self):
         data = {"type": "ide"}
-        template = UITemplate.parse_obj(
+        template = UITemplate.model_validate(
             {
                 "type": "template",
                 "name": "t",
@@ -43,7 +43,7 @@ def test_parses_ide_parameter(self):
 
     def test_parses_resources_parameter(self):
         data = {"type": "resources"}
-        template = UITemplate.parse_obj(
+        template = UITemplate.model_validate(
             {
                 "type": "template",
                 "name": "t",
@@ -56,7 +56,7 @@ def test_parses_resources_parameter(self):
 
     def test_parses_python_or_docker_parameter(self):
         data = {"type": "python_or_docker"}
-        template = UITemplate.parse_obj(
+        template = UITemplate.model_validate(
             {
                 "type": "template",
                 "name": "t",
@@ -69,7 +69,7 @@ def test_parses_python_or_docker_parameter(self):
 
     def test_parses_repo_parameter(self):
         data = {"type": "repo"}
-        template = UITemplate.parse_obj(
+        template = UITemplate.model_validate(
             {
                 "type": "template",
                 "name": "t",
@@ -82,7 +82,7 @@ def test_parses_repo_parameter(self):
 
     def test_parses_working_dir_parameter(self):
         data = {"type": "working_dir"}
-        template = UITemplate.parse_obj(
+        template = UITemplate.model_validate(
             {
                 "type": "template",
                 "name": "t",
@@ -100,7 +100,7 @@ def test_parses_env_parameter_with_all_fields(self):
             "name": "PASSWORD",
             "value": "$random-password",
         }
-        template = UITemplate.parse_obj(
+        template = UITemplate.model_validate(
             {
                 "type": "template",
                 "name": "t",
@@ -117,7 +117,7 @@ def test_parses_env_parameter_with_all_fields(self):
 
     def test_parses_env_parameter_with_no_optional_fields(self):
         data = {"type": "env"}
-        template = UITemplate.parse_obj(
+        template = UITemplate.model_validate(
             {
                 "type": "template",
                 "name": "t",
@@ -135,7 +135,7 @@ def test_parses_env_parameter_with_no_optional_fields(self):
     def test_rejects_unknown_parameter_type(self):
         data = {"type": "unknown_type"}
         with pytest.raises(ValidationError):
-            UITemplate.parse_obj(
+            UITemplate.model_validate(
                 {
                     "type": "template",
                     "name": "t",
@@ -163,7 +163,7 @@ def test_parses_desktop_ide_template(self):
             ],
             "configuration": {"type": "dev-environment"},
         }
-        template = UITemplate.parse_obj(data)
+        template = UITemplate.model_validate(data)
         assert template.name == "desktop-ide"
         assert template.title == "Desktop IDE"
         assert (
@@ -202,7 +202,7 @@ def test_parses_web_based_ide_template(self):
                 "probes": [{"type": "http", "url": "/healthz"}],
             },
         }
-        template = UITemplate.parse_obj(data)
+        template = UITemplate.model_validate(data)
         assert template.name == "in-browser-ide"
         assert template.title == "In-browser IDE"
         assert len(template.parameters) == 6
@@ -212,7 +212,7 @@ def test_parses_web_based_ide_template(self):
 
     def test_rejects_wrong_type(self):
         with pytest.raises(ValidationError):
-            UITemplate.parse_obj(
+            UITemplate.model_validate(
                 {
                     "type": "not-a-template",
                     "name": "t",
@@ -223,7 +223,7 @@ def test_rejects_wrong_type(self):
 
     def test_rejects_missing_configuration(self):
         with pytest.raises(ValidationError):
-            UITemplate.parse_obj(
+            UITemplate.model_validate(
                 {
                     "type": "template",
                     "name": "t",
@@ -233,7 +233,7 @@ def test_rejects_missing_configuration(self):
 
     def test_rejects_missing_name(self):
         with pytest.raises(ValidationError):
-            UITemplate.parse_obj(
+            UITemplate.model_validate(
                 {
                     "type": "template",
                     "title": "T",
@@ -242,7 +242,7 @@ def test_rejects_missing_name(self):
             )
 
     def test_empty_parameters_default(self):
-        template = UITemplate.parse_obj(
+        template = UITemplate.model_validate(
             {
                 "type": "template",
                 "name": "t",
@@ -253,7 +253,7 @@ def test_empty_parameters_default(self):
         assert template.parameters == []
 
     def test_description_is_optional(self):
-        template = UITemplate.parse_obj(
+        template = UITemplate.model_validate(
             {
                 "type": "template",
                 "name": "t",
diff --git a/src/tests/_internal/core/models/test_volumes.py b/src/tests/_internal/core/models/test_volumes.py
index 3fddfb92fd..9f95b1bffe 100644
--- a/src/tests/_internal/core/models/test_volumes.py
+++ b/src/tests/_internal/core/models/test_volumes.py
@@ -1,5 +1,5 @@
 import pytest
-from pydantic import ValidationError, parse_obj_as
+from pydantic import ValidationError
 
 from dstack._internal.core.models.volumes import (
     InstanceMountPoint,
@@ -15,8 +15,8 @@ def test_parse(self):
         )
 
     def test_path_normalization(self):
-        assert parse_obj_as(
-            VolumeMountPoint, {"name": "my-vol", "path": "/path/./to///dir/"}
+        assert VolumeMountPoint.model_validate(
+            {"name": "my-vol", "path": "/path/./to///dir/"}
         ) == VolumeMountPoint(name="my-vol", path="/path/to/dir")
 
     @pytest.mark.parametrize("value", ["my-vol", "my-vol:/run:ro"])
@@ -26,15 +26,15 @@ def test_parse_error_invalid_format(self, value: str):
 
     def test_validation_error_empty_path(self):
         with pytest.raises(ValidationError, match="empty path"):
-            parse_obj_as(VolumeMountPoint, {"name": "vol", "path": ""})
+            VolumeMountPoint.model_validate({"name": "vol", "path": ""})
 
     def test_validation_error_rel_path(self):
         with pytest.raises(ValidationError, match="path must be absolute"):
-            parse_obj_as(VolumeMountPoint, {"name": "vol", "path": "rel/path"})
+            VolumeMountPoint.model_validate({"name": "vol", "path": "rel/path"})
 
     def test_validation_error_parent_dir(self):
         with pytest.raises(ValidationError, match=r"\.\. are not allowed"):
-            parse_obj_as(VolumeMountPoint, {"name": "vol", "path": "/path/../to"})
+            VolumeMountPoint.model_validate({"name": "vol", "path": "/path/../to"})
 
 
 class TestInstanceBindMountPoint:
@@ -44,8 +44,8 @@ def test_parse(self):
         )
 
     def test_path_normalization(self):
-        assert parse_obj_as(
-            InstanceMountPoint, {"instance_path": "/host/.//path/", "path": "/run//./path"}
+        assert InstanceMountPoint.model_validate(
+            {"instance_path": "/host/.//path/", "path": "/run//./path"}
         ) == InstanceMountPoint(instance_path="/host/path", path="/run/path")
 
     @pytest.mark.parametrize("value", ["/path", "/host/path:/run/path:ro"])
@@ -58,21 +58,21 @@ def test_validation_error_empty_path(self, field: str):
         data = {"instance_path": "/instance_path", "path": "/run_path"}
         data[field] = ""
         with pytest.raises(ValidationError, match="empty path"):
-            parse_obj_as(InstanceMountPoint, data)
+            InstanceMountPoint.model_validate(data)
 
     @pytest.mark.parametrize("field", ["instance_path", "path"])
     def test_validation_error_rel_path(self, field: str):
         data = {"instance_path": "/instance_path", "path": "/run_path"}
         data[field] = "./rel/path"
         with pytest.raises(ValidationError, match="path must be absolute"):
-            parse_obj_as(InstanceMountPoint, data)
+            InstanceMountPoint.model_validate(data)
 
     @pytest.mark.parametrize("field", ["instance_path", "path"])
     def test_validation_error_parent_dir(self, field: str):
         data = {"instance_path": "/instance_path", "path": "/run_path"}
         data[field] = "/path/../to"
         with pytest.raises(ValidationError, match=r"\.\. are not allowed"):
-            parse_obj_as(InstanceMountPoint, data)
+            InstanceMountPoint.model_validate(data)
 
 
 class TestParseMountPoint:
diff --git a/src/tests/_internal/core/services/test_diff.py b/src/tests/_internal/core/services/test_diff.py
index 7b243f8a63..097961f96d 100644
--- a/src/tests/_internal/core/services/test_diff.py
+++ b/src/tests/_internal/core/services/test_diff.py
@@ -58,36 +58,6 @@ class _CoreModelAB(_CoreModelA, _CoreModelB):
                 {},
                 id="core-model-no-diff",
             ),
-            pytest.param(
-                _CoreModelA.__request__(a=1, b="x"),
-                _CoreModelA.__request__(a=1, b="y"),
-                {"b": ModelFieldDiff(old="x", new="y")},
-                id="core-model-request",
-            ),
-            pytest.param(
-                _CoreModelA.__response__(a=1, b="x"),
-                _CoreModelA.__response__(a=1, b="y"),
-                {"b": ModelFieldDiff(old="x", new="y")},
-                id="core-model-response",
-            ),
-            pytest.param(
-                _CoreModelA.__request__(a=1, b="x"),
-                _CoreModelA.__response__(a=1, b="y"),
-                {"b": ModelFieldDiff(old="x", new="y")},
-                id="core-model-request-response",
-            ),
-            pytest.param(
-                _CoreModelA(a=1, b="x"),
-                _CoreModelA.__request__(a=1, b="y"),
-                {"b": ModelFieldDiff(old="x", new="y")},
-                id="core-model-base-request",
-            ),
-            pytest.param(
-                _CoreModelA(a=1, b="x"),
-                _CoreModelA.__response__(a=1, b="y"),
-                {"b": ModelFieldDiff(old="x", new="y")},
-                id="core-model-base-response",
-            ),
         ],
     )
     def test_diff_models(self, old: BaseModel, new: BaseModel, expected: ModelDiff) -> None:
diff --git a/src/tests/_internal/pydantic_compat/backend_factories.py b/src/tests/_internal/pydantic_compat/backend_factories.py
index 7cd16ef9a2..db76c39469 100644
--- a/src/tests/_internal/pydantic_compat/backend_factories.py
+++ b/src/tests/_internal/pydantic_compat/backend_factories.py
@@ -6,10 +6,10 @@
 
 Three `Text` columns are involved, one registry each:
 
-- `BackendModel.config` holds `XStoredConfig(...).json()`, read back as the splice
-  `XConfig(**json.loads(config), creds=XCreds.parse_raw(auth))`. The registries below keep the
+- `BackendModel.config` holds `XStoredConfig(...).model_dump_json()`, read back as the splice
+  `XConfig(**json.loads(config), creds=XCreds.model_validate_json(auth))`. The registries below keep the
   two halves apart the way the columns do, so a fixture matches one column's bytes exactly.
-- `BackendModel.auth` holds `XCreds(...).json()`.
+- `BackendModel.auth` holds `XCreds(...).model_dump_json()`.
 - `InstanceModel.backend_data` and `VolumeModel.backend_data` hold a `*BackendData` blob.
 """
 
@@ -327,17 +327,17 @@ def vultr_stored_config() -> VultrStoredConfig:
 
 
 def aws_creds_access_key() -> AWSCreds:
-    return AWSCreds.parse_obj(
+    return AWSCreds.model_validate(
         {"type": "access_key", "access_key": "AKIAIOSFODNN7EXAMPLE", "secret_key": "wJalrXUtnFEMI"}
     )
 
 
 def aws_creds_default() -> AWSCreds:
-    return AWSCreds.parse_obj({"type": "default"})
+    return AWSCreds.model_validate({"type": "default"})
 
 
 def azure_creds_client() -> AzureCreds:
-    return AzureCreds.parse_obj(
+    return AzureCreds.model_validate(
         {
             "type": "client",
             "client_id": "aaaaaaaa-bbbb-cccc-dddd-eeeeeeeeeeee",
@@ -348,21 +348,21 @@ def azure_creds_client() -> AzureCreds:
 
 
 def azure_creds_default() -> AzureCreds:
-    return AzureCreds.parse_obj({"type": "default"})
+    return AzureCreds.model_validate({"type": "default"})
 
 
 def gcp_creds_service_account() -> GCPCreds:
-    return GCPCreds.parse_obj(
+    return GCPCreds.model_validate(
         {"type": "service_account", "filename": "", "data": _SERVICE_ACCOUNT_JSON}
     )
 
 
 def gcp_creds_default() -> GCPCreds:
-    return GCPCreds.parse_obj({"type": "default"})
+    return GCPCreds.model_validate({"type": "default"})
 
 
 def oci_creds_client() -> OCICreds:
-    return OCICreds.parse_obj(
+    return OCICreds.model_validate(
         {
             "type": "client",
             "user": "ocid1.user.oc1..aaaaaaaadstack",
@@ -375,7 +375,7 @@ def oci_creds_client() -> OCICreds:
 
 
 def oci_creds_default() -> OCICreds:
-    return OCICreds.parse_obj({"type": "default"})
+    return OCICreds.model_validate({"type": "default"})
 
 
 def cloudrift_creds() -> CloudRiftCreds:
diff --git a/src/tests/_internal/pydantic_compat/compare.py b/src/tests/_internal/pydantic_compat/compare.py
index 6096f6655d..0f69dc9abd 100644
--- a/src/tests/_internal/pydantic_compat/compare.py
+++ b/src/tests/_internal/pydantic_compat/compare.py
@@ -110,19 +110,11 @@ def type_map(value: Any, path: str = "", out: Union[dict, None] = None) -> dict:
 
 def class_name(value: Any) -> str:
     """
-    The model's name with pydantic-duality's generated suffix removed.
+    The model's class name.
 
-    Duality names its concrete classes `XRequest` / `XResponse`, and those suffixes vanish in v2,
-    so leaving them in would make every line of every type map diff on the migration branch.
-
-    Strip only for classes duality actually generated, which is what `__response__` identifies.
-    Plenty of models are genuinely *named* `...Request` — every gateway registry schema, for one —
-    and those are plain `BaseModel`, so stripping there would report `RegisterService` for a class
-    called `RegisterServiceRequest`.
+    Kept as a function because the fixtures were generated under pydantic v1, where this stripped
+    the `Request`/`Response` suffix pydantic-duality gave its generated classes. A model is a single
+    class now, so there is nothing to strip and those fixtures still match.
     """
     cls = value if isinstance(value, type) else type(value)
-    name = cls.__name__
-    if hasattr(cls, "__response__"):
-        for suffix in ("Request", "Response"):
-            name = name.removesuffix(suffix)
-    return name
+    return cls.__name__
diff --git a/src/tests/_internal/pydantic_compat/compat.py b/src/tests/_internal/pydantic_compat/compat.py
index 94ccb78f1d..aa24e7296e 100644
--- a/src/tests/_internal/pydantic_compat/compat.py
+++ b/src/tests/_internal/pydantic_compat/compat.py
@@ -1,9 +1,10 @@
 """
-The only place in this package allowed to branch on pydantic version.
+The one place in this package that knows how each `extra` mode is spelled.
 
-Every test here has to run unchanged on v1 and v2, which rules out touching the duality API
-directly. Forbidding extra fields needs no help — `parse_obj` forbids them in both versions.
-Ignoring them does: v1 spells it `X.__response__`, and v2 will spell it `validate_extra_ignore`.
+Every test here was written to run unchanged on v1 and v2, which ruled out touching the duality
+API directly. Forbidding extra fields needs no help — `model_validate` forbids them in both
+versions. Ignoring them did: v1 spelled it with the duality response variant, and v2 spells it
+`validate_extra_ignore`.
 
 Both helpers are named for the `extra` setting they apply, deliberately avoiding the word
 "strict": pydantic's `strict` is an unrelated axis that turns off type coercion, and a migration
@@ -13,10 +14,9 @@
 
 from typing import Any
 
-import pydantic
 from pydantic import BaseModel
 
-PYDANTIC_V1 = pydantic.VERSION.startswith("1.")
+from dstack._internal.core.models.common import validate_extra_ignore
 
 
 def parse_forbid_extra(model: Any, data: Any) -> BaseModel:
@@ -26,7 +26,7 @@ def parse_forbid_extra(model: Any, data: Any) -> BaseModel:
     An unknown field is an error, which is what makes `dstack apply` report a typo'd key instead
     of silently ignoring it.
     """
-    return model.parse_obj(data)
+    return model.model_validate(data)
 
 
 def parse_ignore_extra(model: Any, data: Any) -> BaseModel:
@@ -36,12 +36,6 @@ def parse_ignore_extra(model: Any, data: Any) -> BaseModel:
     Unknown fields are dropped, which is what lets an older reader survive a newer writer, so it
     is the behaviour the whole migration has to preserve.
     """
-    if not hasattr(model, "__response__"):
-        # Plain `BaseModel` rather than `CoreModel` — the proxy and gateway schemas. Their default
-        # is already extra="ignore" in both pydantic versions, so `parse_obj` is the ignore path
-        # and there is no duality variant to reach for.
-        return model.parse_obj(data)
-    if PYDANTIC_V1:
-        return model.__response__.parse_obj(data)
-    # v2: from dstack._internal.core.models.common import validate_extra_ignore
-    raise NotImplementedError("wire up validate_extra_ignore when the v2 branch lands")
+    # Works for plain `BaseModel` too (the proxy and gateway schemas, whose default is already
+    # extra="ignore"): the per-call override applies to any model, not just `CoreModel`.
+    return validate_extra_ignore(model, data)
diff --git a/src/tests/_internal/pydantic_compat/factories.py b/src/tests/_internal/pydantic_compat/factories.py
index 2e3c76f8e6..f0a7e991b1 100644
--- a/src/tests/_internal/pydantic_compat/factories.py
+++ b/src/tests/_internal/pydantic_compat/factories.py
@@ -16,6 +16,7 @@
     DevEnvironmentConfiguration,
     PythonVersion,
 )
+from dstack._internal.core.models.duration import Duration
 from dstack._internal.core.models.envs import Env
 from dstack._internal.core.models.fleets import (
     Fleet,
@@ -262,7 +263,10 @@ def job_spec() -> JobSpec:
         job_num=0,
         job_name="test-run-0-0",
         commands=["/bin/bash", "-i", "-c", "echo hi"],
-        env=Env.parse_obj({"A": "1"}),
+        # `JobSpec.env` is a plain `Dict[str, str]`, not an `Env`. v1 happened to coerce an `Env`
+        # into it via the mapping protocol; v2 requires a real dict, which is what production
+        # passes anyway (`_env()` calls `Env.as_dict()`).
+        env={"A": "1"},
         image_name="dstackai/base:latest",
         requirements=requirements(),
         max_duration=7200,
@@ -287,7 +291,7 @@ def profile() -> Profile:
         instance_types=["p4d.24xlarge"],
         reservation="test-reservation",
         spot_policy=SpotPolicy.AUTO,
-        retry=ProfileRetry(on_events=[RetryEvent.NO_CAPACITY], duration=3600),
+        retry=ProfileRetry(on_events=[RetryEvent.NO_CAPACITY], duration=Duration(3600)),
         max_duration=7200,
         stop_duration=300,
         idle_duration=600,
@@ -325,7 +329,7 @@ def requirements() -> Requirements:
 
 
 def resources() -> Resources:
-    """`Resources.dict()` rewrites `cpu` for old clients — the other custom serializer."""
+    """`Resources.model_dump()` rewrites `cpu` for old clients — the other custom serializer."""
     return Resources(
         cpus=8,
         memory_mib=16384,
@@ -354,7 +358,7 @@ def run_spec() -> RunSpec:
             # `image` is deliberately absent: it is mutually exclusive with `python`, and `python`
             # is the more valuable of the two to pin because it is a str enum fed by a YAML float.
             python=PythonVersion.PY311,
-            env=Env.parse_obj({"HF_TOKEN": "secret"}),
+            env=Env.model_validate({"HF_TOKEN": "secret"}),
             working_dir="/workflow",
             inactivity_duration=3600,
             resources=ResourcesSpec(
@@ -384,7 +388,7 @@ def volume_provisioning_data() -> VolumeProvisioningData:
 
 
 # --- API responses -------------------------------------------------------------------
-# Returned from a router through `CustomORJSONResponse`. Chosen by greedy set cover so that
+# Returned from a router through `CustomJSONResponse`. Chosen by greedy set cover so that
 # between them they reach every model class reachable from any response model — 129 of 129.
 # `run` and `instance` are absent on purpose: `run_plan` and `fleet` already reach everything
 # they would add.
@@ -392,7 +396,7 @@ def volume_provisioning_data() -> VolumeProvisioningData:
 
 def fleet() -> Fleet:
     """
-    The default `FleetNodesSpec` has `target == min`, which is what makes `FleetNodesSpec.dict()`
+    The default `FleetNodesSpec` has `target == min`, which is what makes `FleetNodesSpec.model_dump()`
     drop `target` — the old-client compat hack from #3066. That override becomes a
     `@model_serializer` in v2, so this fixture is what proves the hack survived.
     """
diff --git a/src/tests/_internal/pydantic_compat/fixtures/parsing/api_request/apply_run_plan_request.types.json b/src/tests/_internal/pydantic_compat/fixtures/parsing/api_request/apply_run_plan_request.types.json
index 50646cd45b..c3225e8f74 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/parsing/api_request/apply_run_plan_request.types.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/parsing/api_request/apply_run_plan_request.types.json
@@ -4,8 +4,7 @@
   "/plan/run_spec": "RunSpec",
   "/plan/run_spec/configuration": "TaskConfiguration",
   "/plan/run_spec/configuration/env": "Env",
-  "/plan/run_spec/configuration/env/__root__/HF_TOKEN": "EnvSentinel",
-  "/plan/run_spec/configuration/max_duration": "Duration",
+  "/plan/run_spec/configuration/env/root/HF_TOKEN": "EnvSentinel",
   "/plan/run_spec/configuration/python": "PythonVersion",
   "/plan/run_spec/configuration/resources": "ResourcesSpec",
   "/plan/run_spec/configuration/resources/cpu": "CPUSpec",
@@ -20,11 +19,9 @@
   "/plan/run_spec/merged_profile": "Profile",
   "/plan/run_spec/merged_profile/backends/0": "BackendType",
   "/plan/run_spec/merged_profile/creation_policy": "CreationPolicy",
-  "/plan/run_spec/merged_profile/max_duration": "Duration",
   "/plan/run_spec/merged_profile/spot_policy": "SpotPolicy",
   "/plan/run_spec/profile": "Profile",
   "/plan/run_spec/profile/backends/0": "BackendType",
-  "/plan/run_spec/profile/max_duration": "Duration",
   "/plan/run_spec/profile/spot_policy": "SpotPolicy",
   "/plan/run_spec/repo_data": "RemoteRunRepoData"
 }
diff --git a/src/tests/_internal/pydantic_compat/fixtures/parsing/api_response/fleet.values.json b/src/tests/_internal/pydantic_compat/fixtures/parsing/api_response/fleet.values.json
index fc93513d22..f044a57c7e 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/parsing/api_response/fleet.values.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/parsing/api_response/fleet.values.json
@@ -1,5 +1,5 @@
 {
-  "created_at": "2024-01-02T03:04:05+00:00",
+  "created_at": "2024-01-02T03:04:05Z",
   "id": "11111111-1111-1111-1111-111111111111",
   "instances": [],
   "name": "test-fleet",
diff --git a/src/tests/_internal/pydantic_compat/fixtures/parsing/api_response/gateway.values.json b/src/tests/_internal/pydantic_compat/fixtures/parsing/api_response/gateway.values.json
index 0e2961b328..0364f1826c 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/parsing/api_response/gateway.values.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/parsing/api_response/gateway.values.json
@@ -16,7 +16,7 @@
     "tags": null,
     "type": "gateway"
   },
-  "created_at": "2025-03-14T09:26:53+00:00",
+  "created_at": "2025-03-14T09:26:53Z",
   "default": true,
   "hostname": "gateway.inference.example.com",
   "id": "b41d9e77-3c8a-4f21-9d6e-5a7b8c9d0e1f",
diff --git a/src/tests/_internal/pydantic_compat/fixtures/parsing/api_response/project.values.json b/src/tests/_internal/pydantic_compat/fixtures/parsing/api_response/project.values.json
index 60e523ab83..63f79ddfcb 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/parsing/api_response/project.values.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/parsing/api_response/project.values.json
@@ -44,7 +44,7 @@
   ],
   "owner": {
     "active": true,
-    "created_at": "2025-01-08T11:02:41+00:00",
+    "created_at": "2025-01-08T11:02:41Z",
     "email": "alice@example.com",
     "global_role": "admin",
     "id": "9c2f7b31-5e4a-4d8c-9f01-2b3c4d5e6f70",
diff --git a/src/tests/_internal/pydantic_compat/fixtures/parsing/api_response/user_with_creds.values.json b/src/tests/_internal/pydantic_compat/fixtures/parsing/api_response/user_with_creds.values.json
index c2778861e1..60230fa2eb 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/parsing/api_response/user_with_creds.values.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/parsing/api_response/user_with_creds.values.json
@@ -1,6 +1,6 @@
 {
   "active": true,
-  "created_at": "2025-03-14T09:26:53+00:00",
+  "created_at": "2025-03-14T09:26:53Z",
   "creds": {
     "token": "0a1b2c3d-4e5f-6071-8293-a4b5c6d7e8f9"
   },
diff --git a/src/tests/_internal/pydantic_compat/fixtures/parsing/api_response/volume.values.json b/src/tests/_internal/pydantic_compat/fixtures/parsing/api_response/volume.values.json
index 5a3344d2eb..e28aa99aa2 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/parsing/api_response/volume.values.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/parsing/api_response/volume.values.json
@@ -13,12 +13,12 @@
     "volume_id": null
   },
   "cost": 42.5,
-  "created_at": "2025-04-02T16:41:07+00:00",
+  "created_at": "2025-04-02T16:41:07Z",
   "deleted": false,
   "deleted_at": null,
   "external": false,
   "id": "e7f80912-3a4b-4c5d-9e6f-708192a3b4c5",
-  "last_processed_at": "2025-04-02T16:41:22+00:00",
+  "last_processed_at": "2025-04-02T16:41:22Z",
   "name": "training-data",
   "project_name": "main",
   "provisioning_data": {
diff --git a/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/aws.access_key.types.json b/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/aws.access_key.types.json
index ed38da62cb..0a458d43c1 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/aws.access_key.types.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/aws.access_key.types.json
@@ -1,4 +1,4 @@
 {
   "/": "AWSCreds",
-  "/__root__": "AWSAccessKeyCreds"
+  "/root": "AWSAccessKeyCreds"
 }
diff --git a/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/aws.default.types.json b/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/aws.default.types.json
index 361c6cc505..fa74a50d72 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/aws.default.types.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/aws.default.types.json
@@ -1,4 +1,4 @@
 {
   "/": "AWSCreds",
-  "/__root__": "AWSDefaultCreds"
+  "/root": "AWSDefaultCreds"
 }
diff --git a/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/azure.client.types.json b/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/azure.client.types.json
index c5fd8d1a97..1142a9d291 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/azure.client.types.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/azure.client.types.json
@@ -1,4 +1,4 @@
 {
   "/": "AzureCreds",
-  "/__root__": "AzureClientCreds"
+  "/root": "AzureClientCreds"
 }
diff --git a/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/azure.default.types.json b/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/azure.default.types.json
index 387b7147bd..83b234f773 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/azure.default.types.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/azure.default.types.json
@@ -1,4 +1,4 @@
 {
   "/": "AzureCreds",
-  "/__root__": "AzureDefaultCreds"
+  "/root": "AzureDefaultCreds"
 }
diff --git a/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/gcp.default.types.json b/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/gcp.default.types.json
index b51f633e4c..f32a017c54 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/gcp.default.types.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/gcp.default.types.json
@@ -1,4 +1,4 @@
 {
   "/": "GCPCreds",
-  "/__root__": "GCPDefaultCreds"
+  "/root": "GCPDefaultCreds"
 }
diff --git a/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/gcp.service_account.types.json b/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/gcp.service_account.types.json
index 5225a9485c..d211559299 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/gcp.service_account.types.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/gcp.service_account.types.json
@@ -1,4 +1,4 @@
 {
   "/": "GCPCreds",
-  "/__root__": "GCPServiceAccountCreds"
+  "/root": "GCPServiceAccountCreds"
 }
diff --git a/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/oci.client.types.json b/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/oci.client.types.json
index 2ea26b6f8b..47adf538ae 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/oci.client.types.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/oci.client.types.json
@@ -1,4 +1,4 @@
 {
   "/": "OCICreds",
-  "/__root__": "OCIClientCreds"
+  "/root": "OCIClientCreds"
 }
diff --git a/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/oci.default.types.json b/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/oci.default.types.json
index 6f460a30c0..a12b71aea8 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/oci.default.types.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/parsing/backend_creds/oci.default.types.json
@@ -1,4 +1,4 @@
 {
   "/": "OCICreds",
-  "/__root__": "OCIDefaultCreds"
+  "/root": "OCIDefaultCreds"
 }
diff --git a/src/tests/_internal/pydantic_compat/fixtures/parsing/config/dev_environment.types.json b/src/tests/_internal/pydantic_compat/fixtures/parsing/config/dev_environment.types.json
index 4d2dbc17ff..1416a0625f 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/parsing/config/dev_environment.types.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/parsing/config/dev_environment.types.json
@@ -1,7 +1,6 @@
 {
   "/": "DevEnvironmentConfiguration",
   "/env": "Env",
-  "/inactivity_duration": "Duration",
   "/python": "PythonVersion",
   "/resources": "ResourcesSpec",
   "/resources/cpu": "CPUSpec",
diff --git a/src/tests/_internal/pydantic_compat/fixtures/parsing/config/fleet.types.json b/src/tests/_internal/pydantic_compat/fixtures/parsing/config/fleet.types.json
index 8887ca1483..484d864ada 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/parsing/config/fleet.types.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/parsing/config/fleet.types.json
@@ -1,7 +1,6 @@
 {
   "/": "FleetConfiguration",
   "/env": "Env",
-  "/idle_duration": "Duration",
   "/nodes": "FleetNodesSpec",
   "/resources": "ResourcesSpec",
   "/resources/cpu": "CPUSpec",
diff --git a/src/tests/_internal/pydantic_compat/fixtures/parsing/config/profiles.types.json b/src/tests/_internal/pydantic_compat/fixtures/parsing/config/profiles.types.json
index 93d04a9e6d..b349fc4668 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/parsing/config/profiles.types.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/parsing/config/profiles.types.json
@@ -1,7 +1,5 @@
 {
   "/": "ProfilesConfig",
   "/profiles/0": "Profile",
-  "/profiles/0/idle_duration": "Duration",
-  "/profiles/0/max_duration": "Duration",
   "/profiles/0/spot_policy": "SpotPolicy"
 }
diff --git a/src/tests/_internal/pydantic_compat/fixtures/parsing/config/task.types.json b/src/tests/_internal/pydantic_compat/fixtures/parsing/config/task.types.json
index a29b85f05f..7420ad346b 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/parsing/config/task.types.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/parsing/config/task.types.json
@@ -1,8 +1,7 @@
 {
   "/": "TaskConfiguration",
   "/env": "Env",
-  "/env/__root__/B": "EnvSentinel",
-  "/max_duration": "Duration",
+  "/env/root/B": "EnvSentinel",
   "/python": "PythonVersion",
   "/resources": "ResourcesSpec",
   "/resources/cpu": "CPUSpec",
diff --git a/src/tests/_internal/pydantic_compat/fixtures/parsing/db/fleet_spec.types.json b/src/tests/_internal/pydantic_compat/fixtures/parsing/db/fleet_spec.types.json
index d24311bc90..20e1d318be 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/parsing/db/fleet_spec.types.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/parsing/db/fleet_spec.types.json
@@ -3,7 +3,6 @@
   "/configuration": "FleetConfiguration",
   "/configuration/backends/0": "BackendType",
   "/configuration/env": "Env",
-  "/configuration/idle_duration": "Duration",
   "/configuration/nodes": "FleetNodesSpec",
   "/configuration/placement": "InstanceGroupPlacement",
   "/configuration/resources": "ResourcesSpec",
@@ -19,9 +18,7 @@
   "/configuration/spot_policy": "SpotPolicy",
   "/merged_profile": "Profile",
   "/merged_profile/backends/0": "BackendType",
-  "/merged_profile/idle_duration": "Duration",
   "/merged_profile/spot_policy": "SpotPolicy",
   "/profile": "Profile",
-  "/profile/idle_duration": "Duration",
   "/profile/spot_policy": "SpotPolicy"
 }
diff --git a/src/tests/_internal/pydantic_compat/fixtures/parsing/db/fleet_spec.values.json b/src/tests/_internal/pydantic_compat/fixtures/parsing/db/fleet_spec.values.json
index 18cf830b07..da0793eb45 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/parsing/db/fleet_spec.values.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/parsing/db/fleet_spec.values.json
@@ -26,7 +26,7 @@
     "resources": {
       "cpu": {
         "max": null,
-        "min": 2
+        "min": 8
       },
       "disk": {
         "size": {
diff --git a/src/tests/_internal/pydantic_compat/fixtures/parsing/db/job_provisioning_data.values.json b/src/tests/_internal/pydantic_compat/fixtures/parsing/db/job_provisioning_data.values.json
index 46b4cac601..e0505f98d8 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/parsing/db/job_provisioning_data.values.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/parsing/db/job_provisioning_data.values.json
@@ -4,6 +4,7 @@
   "backend_data": "{\"boot_disk_id\": \"vol-0fe1a2b3c4d5e6f78\"}",
   "base_backend": null,
   "dockerized": true,
+  "gpu_driver": null,
   "hostname": "54.221.13.207",
   "instance_id": "i-0abc123def4567890",
   "instance_network": null,
diff --git a/src/tests/_internal/pydantic_compat/fixtures/parsing/db/profile.types.json b/src/tests/_internal/pydantic_compat/fixtures/parsing/db/profile.types.json
index d8193ccef6..56e82a12d5 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/parsing/db/profile.types.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/parsing/db/profile.types.json
@@ -3,8 +3,6 @@
   "/backends/0": "BackendType",
   "/backends/1": "BackendType",
   "/creation_policy": "CreationPolicy",
-  "/idle_duration": "Duration",
-  "/max_duration": "Duration",
   "/retry": "ProfileRetry",
   "/retry/duration": "Duration",
   "/retry/on_events/0": "RetryEvent",
diff --git a/src/tests/_internal/pydantic_compat/fixtures/parsing/db/run_spec.types.json b/src/tests/_internal/pydantic_compat/fixtures/parsing/db/run_spec.types.json
index e11749d6c6..2a6ae08338 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/parsing/db/run_spec.types.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/parsing/db/run_spec.types.json
@@ -2,7 +2,7 @@
   "/": "RunSpec",
   "/configuration": "TaskConfiguration",
   "/configuration/env": "Env",
-  "/configuration/env/__root__/B": "EnvSentinel",
+  "/configuration/env/root/B": "EnvSentinel",
   "/configuration/python": "PythonVersion",
   "/configuration/resources": "ResourcesSpec",
   "/configuration/resources/cpu": "CPUSpec",
@@ -16,8 +16,6 @@
   "/configuration/resources/memory/min": "Memory",
   "/merged_profile": "Profile",
   "/merged_profile/creation_policy": "CreationPolicy",
-  "/merged_profile/max_duration": "Duration",
   "/profile": "Profile",
-  "/profile/max_duration": "Duration",
   "/repo_data": "LocalRunRepoData"
 }
diff --git a/src/tests/_internal/pydantic_compat/fixtures/schema/configuration.json b/src/tests/_internal/pydantic_compat/fixtures/schema/configuration.json
index 8d1ac83a94..a888ed2931 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/schema/configuration.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/schema/configuration.json
@@ -1,8 +1,6 @@
 {
-  "$schema": "http://json-schema.org/draft-07/schema#",
-  "additionalProperties": true,
-  "definitions": {
-    "ACMGatewayCertificateRequest": {
+  "$defs": {
+    "ACMGatewayCertificate": {
       "additionalProperties": false,
       "properties": {
         "arn": {
@@ -11,11 +9,9 @@
           "type": "string"
         },
         "type": {
+          "const": "acm",
           "default": "acm",
           "description": "Certificates by AWS Certificate Manager (ACM)",
-          "enum": [
-            "acm"
-          ],
           "title": "Type",
           "type": "string"
         }
@@ -23,42 +19,67 @@
       "required": [
         "arn"
       ],
-      "title": "ACMGatewayCertificateRequest",
+      "title": "ACMGatewayCertificate",
       "type": "object"
     },
-    "AWSVolumeConfigurationRequest": {
+    "AWSVolumeConfiguration": {
       "additionalProperties": false,
       "properties": {
         "auto_cleanup_duration": {
           "anyOf": [
+            {
+              "const": "off",
+              "type": "string"
+            },
             {
               "type": "integer"
             },
             {
               "type": "string"
+            },
+            {
+              "type": "boolean"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "Time to wait after volume is no longer used by any job before deleting it. Defaults to keep the volume indefinitely. Use the value `off` or `-1` to disable auto-cleanup",
           "title": "Auto Cleanup Duration"
         },
         "availability_zone": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The volume availability zone",
-          "title": "Availability Zone",
-          "type": "string"
+          "title": "Availability Zone"
         },
         "backend": {
+          "const": "aws",
           "default": "aws",
           "description": "The volume backend",
-          "enum": [
-            "aws"
-          ],
           "title": "Backend",
           "type": "string"
         },
         "name": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The volume name",
-          "title": "Name",
-          "type": "string"
+          "title": "Name"
         },
         "region": {
           "description": "The volume region",
@@ -66,40 +87,67 @@
           "type": "string"
         },
         "size": {
+          "anyOf": [
+            {
+              "type": "number"
+            },
+            {
+              "type": "integer"
+            },
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The volume size. Must be specified when creating new volumes",
-          "title": "Size",
-          "type": "number"
+          "title": "Size"
         },
         "tags": {
-          "additionalProperties": {
-            "type": "string"
-          },
+          "anyOf": [
+            {
+              "additionalProperties": {
+                "type": "string"
+              },
+              "type": "object"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The custom tags to associate with the volume. The tags are also propagated to the underlying backend resources. If there is a conflict with backend-level tags, does not override them",
-          "title": "Tags",
-          "type": "object"
+          "title": "Tags"
         },
         "type": {
+          "const": "volume",
           "default": "volume",
-          "enum": [
-            "volume"
-          ],
           "title": "Type",
           "type": "string"
         },
         "volume_id": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The volume ID. Must be specified when registering external volumes",
-          "title": "Volume Id",
-          "type": "string"
+          "title": "Volume Id"
         }
       },
       "required": [
         "region"
       ],
-      "title": "AWSVolumeConfigurationRequest",
+      "title": "AWSVolumeConfiguration",
       "type": "object"
     },
     "AcceleratorVendor": {
-      "description": "An enumeration.",
       "enum": [
         "nvidia",
         "amd",
@@ -141,7 +189,6 @@
       "type": "string"
     },
     "CPUArchitecture": {
-      "description": "An enumeration.",
       "enum": [
         "x86",
         "arm"
@@ -149,21 +196,25 @@
       "title": "CPUArchitecture",
       "type": "string"
     },
-    "CPUSpecRequest": {
+    "CPUSpec": {
       "additionalProperties": false,
       "properties": {
         "arch": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/CPUArchitecture"
+            },
             {
-              "$ref": "#/definitions/CPUArchitecture"
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The CPU architecture, one of: `x86`, `arm`"
         },
         "count": {
           "anyOf": [
             {
-              "$ref": "#/definitions/Range_int_"
+              "$ref": "#/$defs/Range_int_"
             },
             {
               "type": "integer"
@@ -176,15 +227,13 @@
             "max": null,
             "min": 2
           },
-          "description": "The number of CPU cores",
-          "title": "Count"
+          "description": "The number of CPU cores"
         }
       },
-      "title": "CPUSpecRequest",
+      "title": "CPUSpec",
       "type": "object"
     },
     "CreationPolicy": {
-      "description": "An enumeration.",
       "enum": [
         "reuse",
         "reuse-or-create"
@@ -192,44 +241,81 @@
       "title": "CreationPolicy",
       "type": "string"
     },
-    "DevEnvironmentConfigurationRequest": {
+    "DevEnvironmentConfiguration": {
       "additionalProperties": false,
       "properties": {
         "availability_zones": {
+          "anyOf": [
+            {
+              "items": {
+                "type": "string"
+              },
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The availability zones to consider for provisioning (e.g., `[eu-west-1a, us-west4-a]`)",
-          "items": {
-            "type": "string"
-          },
-          "title": "Availability Zones",
-          "type": "array"
+          "title": "Availability Zones"
         },
         "backend_options": {
+          "anyOf": [
+            {
+              "items": {
+                "$ref": "#/$defs/VastAIProfileOptions"
+              },
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "Backend-specific options, applied only to offers from that backend",
-          "items": {
-            "$ref": "#/definitions/VastAIProfileOptions"
-          },
-          "title": "Backend Options",
-          "type": "array"
+          "title": "Backend Options"
         },
         "backends": {
+          "anyOf": [
+            {
+              "items": {
+                "$ref": "#/$defs/BackendType"
+              },
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The backends to consider for provisioning (e.g., `[aws, gcp]`)",
-          "items": {
-            "$ref": "#/definitions/BackendType"
-          },
-          "type": "array"
+          "title": "Backends"
         },
         "creation_policy": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/CreationPolicy"
+            },
             {
-              "$ref": "#/definitions/CreationPolicy"
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The policy for using instances from fleets: `reuse`, `reuse-or-create`. Defaults to `reuse-or-create`"
         },
         "docker": {
+          "anyOf": [
+            {
+              "type": "boolean"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "Use Docker inside the container. Mutually exclusive with `image`, `python`, and `nvcc`. Overrides `privileged`",
-          "title": "Docker",
-          "type": "boolean"
+          "title": "Docker"
         },
         "dstack": {
           "default": false,
@@ -238,21 +324,21 @@
           "type": "boolean"
         },
         "entrypoint": {
-          "description": "The Docker entrypoint",
-          "title": "Entrypoint",
-          "type": "string"
-        },
-        "env": {
-          "allOf": [
+          "anyOf": [
+            {
+              "type": "string"
+            },
             {
-              "$ref": "#/definitions/Env"
+              "type": "null"
             }
           ],
-          "default": {
-            "__root__": {}
-          },
-          "description": "The mapping or the list of environment variables",
-          "title": "Env"
+          "default": null,
+          "description": "The Docker entrypoint",
+          "title": "Entrypoint"
+        },
+        "env": {
+          "$ref": "#/$defs/Env",
+          "description": "The mapping or the list of environment variables"
         },
         "files": {
           "default": [],
@@ -260,7 +346,7 @@
           "items": {
             "anyOf": [
               {
-                "$ref": "#/definitions/FilePathMappingRequest"
+                "$ref": "#/$defs/FilePathMapping"
               },
               {
                 "type": "string"
@@ -271,19 +357,27 @@
           "type": "array"
         },
         "fleets": {
-          "description": "The fleets considered for reuse. For fleets owned by the current project, specify fleet names. For imported fleets, specify `<project name>/<fleet name>`",
-          "items": {
-            "anyOf": [
-              {
-                "$ref": "#/definitions/EntityReferenceRequest"
+          "anyOf": [
+            {
+              "items": {
+                "anyOf": [
+                  {
+                    "$ref": "#/$defs/EntityReference"
+                  },
+                  {
+                    "type": "string"
+                  }
+                ]
               },
-              {
-                "type": "string"
-              }
-            ]
-          },
-          "title": "Fleets",
-          "type": "array"
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
+          "description": "The fleets considered for reuse. For fleets owned by the current project, specify fleet names. For imported fleets, specify `<project name>/<fleet name>`",
+          "title": "Fleets"
         },
         "home_dir": {
           "default": "/root",
@@ -293,56 +387,69 @@
         "ide": {
           "anyOf": [
             {
-              "enum": [
-                "vscode"
-              ],
+              "const": "vscode",
               "type": "string"
             },
             {
-              "enum": [
-                "cursor"
-              ],
+              "const": "cursor",
               "type": "string"
             },
             {
-              "enum": [
-                "windsurf"
-              ],
+              "const": "windsurf",
               "type": "string"
             },
             {
-              "enum": [
-                "zed"
-              ],
+              "const": "zed",
               "type": "string"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The IDE to pre-install. Supported values include `vscode`, `cursor`, `windsurf`, and `zed`. Defaults to no IDE (SSH only)",
           "title": "Ide"
         },
         "idle_duration": {
           "anyOf": [
+            {
+              "const": "off",
+              "type": "string"
+            },
             {
               "type": "integer"
             },
             {
               "type": "string"
+            },
+            {
+              "type": "boolean"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "Time to wait before terminating idle instances. When the run reuses an existing fleet instance, the fleet's `idle_duration` applies. When the run provisions a new instance, the shorter of the fleet's and run's values is used. Defaults to `5m` for runs and `3d` for fleets. Use `off` for unlimited duration. Only applied for VM-based backends",
           "title": "Idle Duration"
         },
         "image": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The name of the Docker image to run. If no `image` is specified, `dstack` uses an Ubuntu 24.04-based Docker image that comes pre-configured with `uv`, `python`, `pip`, the CUDA 13.0 runtime, InfiniBand, NCCL, and NCCL tests. It may also include provider-specific components such as EFA support on AWS. For non-Nvidia accelerators or NVidia GPUs unsupported by CUDA 13.0 (e.g. V100, P100), specify a custom Docker image.",
-          "title": "Image",
-          "type": "string"
+          "title": "Image"
         },
         "inactivity_duration": {
           "anyOf": [
             {
-              "enum": [
-                "off"
-              ],
+              "const": "off",
               "type": "string"
             },
             {
@@ -353,8 +460,12 @@
             },
             {
               "type": "string"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The maximum amount of time the dev environment can be inactive (e.g., `2h`, `1d`, etc). After it elapses, the dev environment is automatically stopped. Inactivity is defined as the absence of SSH connections to the dev environment, including VS Code connections, `ssh <run name>` shells, and attached `dstack apply` or `dstack attach` commands. Use `off` for unlimited duration. Can be updated in-place. Defaults to `off`",
           "title": "Inactivity Duration"
         },
@@ -368,72 +479,109 @@
           "type": "array"
         },
         "instance_types": {
+          "anyOf": [
+            {
+              "items": {
+                "type": "string"
+              },
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The cloud-specific instance types to consider for provisioning (e.g., `[g6e.24xlarge, n1-standard-4]`)",
-          "items": {
-            "type": "string"
-          },
-          "title": "Instance Types",
-          "type": "array"
+          "title": "Instance Types"
         },
         "instances": {
-          "description": "The specific fleet instances to consider for reuse. Each value can be an instance name string, or an object with `name`, `hostname`, or `fleet` and `instance`. When set, the run is only placed on matching existing instances.",
-          "items": {
-            "anyOf": [
-              {
-                "$ref": "#/definitions/InstanceNameSelectorRequest"
-              },
-              {
-                "$ref": "#/definitions/InstanceHostnameSelectorRequest"
-              },
-              {
-                "$ref": "#/definitions/FleetInstanceSelectorRequest"
+          "anyOf": [
+            {
+              "items": {
+                "anyOf": [
+                  {
+                    "$ref": "#/$defs/InstanceNameSelector"
+                  },
+                  {
+                    "$ref": "#/$defs/InstanceHostnameSelector"
+                  },
+                  {
+                    "$ref": "#/$defs/FleetInstanceSelector"
+                  },
+                  {
+                    "type": "string"
+                  }
+                ]
               },
-              {
-                "minLength": 1,
-                "type": "string"
-              }
-            ]
-          },
-          "minItems": 1,
-          "title": "Instances",
-          "type": "array"
+              "minItems": 1,
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
+          "description": "The specific fleet instances to consider for reuse. Each value can be an instance name string, or an object with `name`, `hostname`, or `fleet` and `instance`. When set, the run is only placed on matching existing instances.",
+          "title": "Instances"
         },
         "max_duration": {
           "anyOf": [
             {
-              "enum": [
-                "off"
-              ],
-              "type": "string"
+              "type": "integer"
             },
             {
-              "type": "integer"
+              "type": "string"
             },
             {
               "type": "boolean"
             },
             {
-              "type": "string"
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The maximum duration of a run (e.g., `2h`, `1d`, etc) in a running state, excluding provisioning and pulling. After it elapses, the run is automatically stopped. Use `off` for unlimited duration. Defaults to `off`",
           "title": "Max Duration"
         },
         "max_price": {
+          "anyOf": [
+            {
+              "exclusiveMinimum": 0.0,
+              "type": "number"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The maximum instance price per hour, in dollars",
-          "exclusiveMinimum": 0.0,
-          "title": "Max Price",
-          "type": "number"
+          "title": "Max Price"
         },
         "name": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The run name. If not specified, a random name is generated",
-          "title": "Name",
-          "type": "string"
+          "title": "Name"
         },
         "nvcc": {
+          "anyOf": [
+            {
+              "type": "boolean"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "Use image with NVIDIA CUDA Compiler (NVCC) included. Mutually exclusive with `image` and `docker`",
-          "title": "Nvcc",
-          "type": "boolean"
+          "title": "Nvcc"
         },
         "ports": {
           "default": [],
@@ -450,7 +598,7 @@
                 "type": "string"
               },
               {
-                "$ref": "#/definitions/PortMappingRequest"
+                "$ref": "#/$defs/PortMapping"
               }
             ]
           },
@@ -458,11 +606,19 @@
           "type": "array"
         },
         "priority": {
-          "description": "The priority of the run, an integer between `0` and `100`. `dstack` tries to provision runs with higher priority first. Defaults to `0`",
-          "maximum": 100,
-          "minimum": 0,
-          "title": "Priority",
-          "type": "integer"
+          "anyOf": [
+            {
+              "maximum": 100,
+              "minimum": 0,
+              "type": "integer"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
+          "description": "The priority of the run, an integer between `0` and `100`. `dstack` tries to provision runs with higher priority first. Defaults to `0`",
+          "title": "Priority"
         },
         "privileged": {
           "default": false,
@@ -471,50 +627,76 @@
           "type": "boolean"
         },
         "python": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/PythonVersion"
+            },
             {
-              "$ref": "#/definitions/PythonVersion"
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The major version of Python. Mutually exclusive with `image` and `docker`"
         },
         "regions": {
+          "anyOf": [
+            {
+              "items": {
+                "type": "string"
+              },
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The regions to consider for provisioning (e.g., `[eu-west-1, us-west4, westeurope]`)",
-          "items": {
-            "type": "string"
-          },
-          "title": "Regions",
-          "type": "array"
+          "title": "Regions"
         },
         "registry_auth": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/RegistryAuth"
+            },
             {
-              "$ref": "#/definitions/RegistryAuthRequest"
+              "type": "null"
             }
           ],
-          "description": "Credentials for pulling a private Docker image",
-          "title": "Registry Auth"
+          "default": null,
+          "description": "Credentials for pulling a private Docker image"
         },
         "repos": {
           "default": [],
           "description": "The list of Git repos",
           "items": {
-            "$ref": "#/definitions/RepoSpecRequest"
+            "anyOf": [
+              {
+                "$ref": "#/$defs/RepoSpec"
+              },
+              {
+                "type": "string"
+              }
+            ]
           },
           "title": "Repos",
           "type": "array"
         },
         "reservation": {
-          "description": "The existing reservation to use for instance provisioning. Supports AWS Capacity Reservations, AWS Capacity Blocks, and GCP reservations",
-          "title": "Reservation",
-          "type": "string"
-        },
-        "resources": {
-          "allOf": [
+          "anyOf": [
             {
-              "$ref": "#/definitions/ResourcesSpecRequest"
+              "type": "string"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
+          "description": "The existing reservation to use for instance provisioning. Supports AWS Capacity Reservations, AWS Capacity Blocks, and GCP reservations",
+          "title": "Reservation"
+        },
+        "resources": {
+          "$ref": "#/$defs/ResourcesSpec",
           "default": {
             "cpu": {
               "max": null,
@@ -543,29 +725,35 @@
             },
             "shm_size": null
           },
-          "description": "The resources requirements to run the configuration",
-          "title": "Resources"
+          "description": "The resources requirements to run the configuration"
         },
         "retry": {
           "anyOf": [
             {
-              "$ref": "#/definitions/ProfileRetryRequest"
+              "$ref": "#/$defs/ProfileRetry"
             },
             {
               "type": "boolean"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The policy for resubmitting the run. Defaults to `false`",
           "title": "Retry"
         },
         "schedule": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/Schedule"
+            },
             {
-              "$ref": "#/definitions/ScheduleRequest"
+              "type": "null"
             }
           ],
-          "description": "The schedule for starting the run at specified time",
-          "title": "Schedule"
+          "default": null,
+          "description": "The schedule for starting the run at specified time"
         },
         "setup": {
           "default": [],
@@ -576,94 +764,145 @@
           "type": "array"
         },
         "shell": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The shell used to run commands. Allowed values are `sh`, `bash`, or an absolute path, e.g., `/usr/bin/zsh`. Defaults to `/bin/sh` if the `image` is specified, `/bin/bash` otherwise",
-          "title": "Shell",
-          "type": "string"
+          "title": "Shell"
         },
         "single_branch": {
+          "anyOf": [
+            {
+              "type": "boolean"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "Whether to clone and track only the current branch or all remote branches. Relevant only when using remote Git repos. Defaults to `false` for dev environments and to `true` for tasks and services",
-          "title": "Single Branch",
-          "type": "boolean"
+          "title": "Single Branch"
         },
         "spot_policy": {
-          "allOf": [
+          "anyOf": [
             {
-              "$ref": "#/definitions/SpotPolicy"
+              "$ref": "#/$defs/SpotPolicy"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The policy for provisioning spot or on-demand instances: `spot`, `on-demand`, `auto`. Defaults to `on-demand`"
         },
         "startup_order": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/StartupOrder"
+            },
             {
-              "$ref": "#/definitions/StartupOrder"
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The order in which master and workers jobs are started: `any`, `master-first`, `workers-first`. Defaults to `any`"
         },
         "stop_criteria": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/StopCriteria"
+            },
             {
-              "$ref": "#/definitions/StopCriteria"
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The criteria determining when a multi-node run should be considered finished: `all-done`, `master-done`. Defaults to `all-done`"
         },
         "stop_duration": {
           "anyOf": [
             {
-              "enum": [
-                "off"
-              ],
-              "type": "string"
+              "type": "integer"
             },
             {
-              "type": "integer"
+              "type": "string"
             },
             {
               "type": "boolean"
             },
             {
-              "type": "string"
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The maximum duration of a run graceful stopping. After it elapses, the run is automatically forced stopped. This includes force detaching volumes used by the run. Use `off` for unlimited duration. Defaults to `5m`",
           "title": "Stop Duration"
         },
         "tags": {
-          "additionalProperties": {
-            "type": "string"
-          },
+          "anyOf": [
+            {
+              "additionalProperties": {
+                "type": "string"
+              },
+              "type": "object"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The custom tags to associate with the resource. The tags are also propagated to the underlying backend resources. If there is a conflict with backend-level tags, does not override them",
-          "title": "Tags",
-          "type": "object"
+          "title": "Tags"
         },
         "type": {
+          "const": "dev-environment",
           "default": "dev-environment",
-          "enum": [
-            "dev-environment"
-          ],
           "title": "Type",
           "type": "string"
         },
         "user": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The user inside the container, `user_name_or_id[:group_name_or_id]` (e.g., `ubuntu`, `1000:1000`). Defaults to the default user from the `image`",
-          "title": "User",
-          "type": "string"
+          "title": "User"
         },
         "utilization_policy": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/UtilizationPolicy"
+            },
             {
-              "$ref": "#/definitions/UtilizationPolicyRequest"
+              "type": "null"
             }
           ],
-          "description": "Run termination policy based on utilization",
-          "title": "Utilization Policy"
+          "default": null,
+          "description": "Run termination policy based on utilization"
         },
         "version": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The version of the IDE. For `windsurf`, the version is in the format `version@commit`",
-          "title": "Version",
-          "type": "string"
+          "title": "Version"
         },
         "volumes": {
           "default": [],
@@ -671,10 +910,10 @@
           "items": {
             "anyOf": [
               {
-                "$ref": "#/definitions/VolumeMountPointRequest"
+                "$ref": "#/$defs/VolumeMountPoint"
               },
               {
-                "$ref": "#/definitions/InstanceMountPointRequest"
+                "$ref": "#/$defs/InstanceMountPoint"
               },
               {
                 "type": "string"
@@ -685,21 +924,29 @@
           "type": "array"
         },
         "working_dir": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The absolute path to the working directory inside the container. Defaults to the `image`'s default working directory",
-          "title": "Working Dir",
-          "type": "string"
+          "title": "Working Dir"
         }
       },
-      "title": "DevEnvironmentConfigurationRequest",
+      "title": "DevEnvironmentConfiguration",
       "type": "object"
     },
-    "DiskSpecRequest": {
+    "DiskSpec": {
       "additionalProperties": false,
       "properties": {
         "size": {
           "anyOf": [
             {
-              "$ref": "#/definitions/Range_Memory_"
+              "$ref": "#/$defs/Range_Memory_"
             },
             {
               "type": "integer"
@@ -708,17 +955,16 @@
               "type": "string"
             }
           ],
-          "description": "Disk size",
-          "title": "Size"
+          "description": "Disk size"
         }
       },
       "required": [
         "size"
       ],
-      "title": "DiskSpecRequest",
+      "title": "DiskSpec",
       "type": "object"
     },
-    "EntityReferenceRequest": {
+    "EntityReference": {
       "additionalProperties": false,
       "description": "Cross-project entity reference.",
       "properties": {
@@ -728,21 +974,30 @@
           "type": "string"
         },
         "project": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The project name. If unspecified, refers to the current project",
-          "title": "Project",
-          "type": "string"
+          "title": "Project"
         }
       },
       "required": [
         "name"
       ],
-      "title": "EntityReferenceRequest",
+      "title": "EntityReference",
       "type": "object"
     },
     "Env": {
       "anyOf": [
         {
           "items": {
+            "pattern": "^([a-zA-Z_][a-zA-Z0-9_]*)(=.*$|$)",
             "type": "string"
           },
           "type": "array"
@@ -754,7 +1009,7 @@
                 "type": "string"
               },
               {
-                "$ref": "#/definitions/EnvSentinelRequest"
+                "$ref": "#/$defs/EnvSentinel"
               }
             ]
           },
@@ -762,10 +1017,10 @@
         }
       ],
       "default": {},
-      "description": "Env represents a mapping of process environment variables, as in environ(7).\nEnvironment values may be omitted, in that case the :class:`EnvSentinel`\nobject is used as a placeholder.\n\nTo create an instance from a `dict[str, str]` or a `list[str]` use pydantic's\n:meth:`BaseModel.parse_obj(dict | list)` method.\n\nNB: this is *NOT* a CoreModel, pydantic-duality, which is used as a base\nfor the CoreModel, doesn't play well with custom root models.",
+      "description": "Env represents a mapping of process environment variables, as in environ(7).\nEnvironment values may be omitted, in that case the :class:`EnvSentinel`\nobject is used as a placeholder.\n\nTo create an instance from a `dict[str, str]` or a `list[str]` use pydantic's\n:meth:`BaseModel.model_validate(dict | list)` method.\n\nNB: this is *NOT* a CoreModel. `extra` is meaningless on a root model, but\n`coerce_numbers_to_str` is not: without it `env: {PORT: 8080}` stops parsing, since\npydantic v2 does not coerce a YAML number to a str implicitly.",
       "title": "Env"
     },
-    "EnvSentinelRequest": {
+    "EnvSentinel": {
       "additionalProperties": false,
       "properties": {
         "key": {
@@ -776,10 +1031,10 @@
       "required": [
         "key"
       ],
-      "title": "EnvSentinelRequest",
+      "title": "EnvSentinel",
       "type": "object"
     },
-    "FilePathMappingRequest": {
+    "FilePathMapping": {
       "additionalProperties": false,
       "properties": {
         "local_path": {
@@ -797,41 +1052,64 @@
         "local_path",
         "path"
       ],
-      "title": "FilePathMappingRequest",
+      "title": "FilePathMapping",
       "type": "object"
     },
-    "FleetConfigurationRequest": {
+    "FleetConfiguration": {
       "additionalProperties": false,
       "properties": {
         "availability_zones": {
+          "anyOf": [
+            {
+              "items": {
+                "type": "string"
+              },
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The availability zones to consider for provisioning (e.g., `[eu-west-1a, us-west4-a]`)",
-          "items": {
-            "type": "string"
-          },
-          "title": "Availability Zones",
-          "type": "array"
+          "title": "Availability Zones"
         },
         "backend_options": {
+          "anyOf": [
+            {
+              "items": {
+                "$ref": "#/$defs/VastAIProfileOptions"
+              },
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "Backend-specific options, applied only to offers from that backend",
-          "items": {
-            "$ref": "#/definitions/VastAIProfileOptions"
-          },
-          "title": "Backend Options",
-          "type": "array"
+          "title": "Backend Options"
         },
         "backends": {
+          "anyOf": [
+            {
+              "items": {
+                "$ref": "#/$defs/BackendType"
+              },
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The backends to consider for provisioning (e.g., `[aws, gcp]`)",
-          "items": {
-            "$ref": "#/definitions/BackendType"
-          },
-          "type": "array"
+          "title": "Backends"
         },
         "blocks": {
           "anyOf": [
             {
-              "enum": [
-                "auto"
-              ],
+              "const": "auto",
               "type": "string"
             },
             {
@@ -844,157 +1122,225 @@
           "title": "Blocks"
         },
         "env": {
-          "allOf": [
-            {
-              "$ref": "#/definitions/Env"
-            }
-          ],
-          "default": {
-            "__root__": {}
-          },
-          "description": "The mapping or the list of environment variables",
-          "title": "Env"
+          "$ref": "#/$defs/Env",
+          "description": "The mapping or the list of environment variables"
         },
         "idle_duration": {
           "anyOf": [
+            {
+              "const": "off",
+              "type": "string"
+            },
             {
               "type": "integer"
             },
             {
               "type": "string"
+            },
+            {
+              "type": "boolean"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "Time to wait before terminating idle instances. Instances are not terminated if the fleet is already at `nodes.min`. Defaults to `5m` for runs and `3d` for fleets. Use `off` for unlimited duration",
           "title": "Idle Duration"
         },
         "instance_types": {
+          "anyOf": [
+            {
+              "items": {
+                "type": "string"
+              },
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The cloud-specific instance types to consider for provisioning (e.g., `[g6e.24xlarge, n1-standard-4]`)",
-          "items": {
-            "type": "string"
-          },
-          "title": "Instance Types",
-          "type": "array"
+          "title": "Instance Types"
         },
         "max_price": {
+          "anyOf": [
+            {
+              "exclusiveMinimum": 0.0,
+              "type": "number"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The maximum instance price per hour, in dollars",
-          "exclusiveMinimum": 0.0,
-          "title": "Max Price",
-          "type": "number"
+          "title": "Max Price"
         },
         "name": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The fleet name",
-          "title": "Name",
-          "type": "string"
+          "title": "Name"
         },
         "nodes": {
           "anyOf": [
             {
-              "$ref": "#/definitions/FleetNodesSpecRequest"
+              "$ref": "#/$defs/FleetNodesSpec"
             },
             {
               "type": "integer"
             },
             {
               "type": "string"
+            },
+            {
+              "type": "null"
             }
           ],
-          "description": "The number of instances",
-          "title": "Nodes"
+          "default": null,
+          "description": "The number of instances"
         },
         "placement": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/InstanceGroupPlacement"
+            },
             {
-              "$ref": "#/definitions/InstanceGroupPlacement"
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The placement of instances: `any` or `cluster`"
         },
         "regions": {
+          "anyOf": [
+            {
+              "items": {
+                "type": "string"
+              },
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The regions to consider for provisioning (e.g., `[eu-west-1, us-west4, westeurope]`)",
-          "items": {
-            "type": "string"
-          },
-          "title": "Regions",
-          "type": "array"
+          "title": "Regions"
         },
         "reservation": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The existing reservation to use for instance provisioning. Supports AWS Capacity Reservations, AWS Capacity Blocks, and GCP reservations",
-          "title": "Reservation",
-          "type": "string"
+          "title": "Reservation"
         },
         "resources": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/ResourcesSpec"
+            },
             {
-              "$ref": "#/definitions/ResourcesSpecRequest"
+              "type": "null"
             }
           ],
-          "description": "The resources requirements",
-          "title": "Resources"
+          "default": null,
+          "description": "The resources requirements"
         },
         "retry": {
           "anyOf": [
             {
-              "$ref": "#/definitions/ProfileRetryRequest"
+              "$ref": "#/$defs/ProfileRetry"
             },
             {
               "type": "boolean"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The policy for provisioning retry. Defaults to `false`",
           "title": "Retry"
         },
         "spot_policy": {
-          "allOf": [
+          "anyOf": [
             {
-              "$ref": "#/definitions/SpotPolicy"
+              "$ref": "#/$defs/SpotPolicy"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The policy for provisioning spot or on-demand instances: `spot`, `on-demand`, `auto`. Defaults to `on-demand`"
         },
         "ssh_config": {
-          "allOf": [
+          "anyOf": [
             {
-              "$ref": "#/definitions/SSHParamsRequest"
-            }
+              "$ref": "#/$defs/SSHParams"
+            },
+            {
+              "type": "null"
+            }
           ],
-          "description": "The parameters for adding instances via SSH",
-          "title": "Ssh Config"
+          "default": null,
+          "description": "The parameters for adding instances via SSH"
         },
         "tags": {
-          "additionalProperties": {
-            "type": "string"
-          },
+          "anyOf": [
+            {
+              "additionalProperties": {
+                "type": "string"
+              },
+              "type": "object"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The custom tags to associate with the resource. The tags are also propagated to the underlying backend resources. If there is a conflict with backend-level tags, does not override them",
-          "title": "Tags",
-          "type": "object"
+          "title": "Tags"
         },
         "type": {
+          "const": "fleet",
           "default": "fleet",
-          "enum": [
-            "fleet"
-          ],
           "title": "Type",
           "type": "string"
         }
       },
-      "title": "FleetConfigurationRequest",
+      "title": "FleetConfiguration",
       "type": "object"
     },
-    "FleetInstanceSelectorRequest": {
+    "FleetInstanceSelector": {
       "additionalProperties": false,
       "properties": {
         "fleet": {
           "anyOf": [
             {
-              "$ref": "#/definitions/EntityReferenceRequest"
+              "$ref": "#/$defs/EntityReference"
             },
             {
-              "minLength": 1,
               "type": "string"
             }
           ],
-          "description": "The fleet reference. For fleets owned by the current project, specify the fleet name. For a fleet from another project, specify `<project name>/<fleet name>` or an object with `project` and `name`.",
-          "title": "Fleet"
+          "description": "The fleet reference. For fleets owned by the current project, specify the fleet name. For a fleet from another project, specify `<project name>/<fleet name>` or an object with `project` and `name`."
         },
         "instance": {
           "description": "The fleet instance number",
@@ -1007,16 +1353,24 @@
         "fleet",
         "instance"
       ],
-      "title": "FleetInstanceSelectorRequest",
+      "title": "FleetInstanceSelector",
       "type": "object"
     },
-    "FleetNodesSpecRequest": {
+    "FleetNodesSpec": {
       "additionalProperties": false,
       "properties": {
         "max": {
+          "anyOf": [
+            {
+              "type": "integer"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The maximum number of instances allowed in the fleet. Unlimited if not specified",
-          "title": "Max",
-          "type": "integer"
+          "title": "Max"
         },
         "min": {
           "description": "The minimum number of instances to maintain in the fleet",
@@ -1033,42 +1387,67 @@
         "min",
         "target"
       ],
-      "title": "FleetNodesSpecRequest",
+      "title": "FleetNodesSpec",
       "type": "object"
     },
-    "GCPVolumeConfigurationRequest": {
+    "GCPVolumeConfiguration": {
       "additionalProperties": false,
       "properties": {
         "auto_cleanup_duration": {
           "anyOf": [
+            {
+              "const": "off",
+              "type": "string"
+            },
             {
               "type": "integer"
             },
             {
               "type": "string"
+            },
+            {
+              "type": "boolean"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "Time to wait after volume is no longer used by any job before deleting it. Defaults to keep the volume indefinitely. Use the value `off` or `-1` to disable auto-cleanup",
           "title": "Auto Cleanup Duration"
         },
         "availability_zone": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The volume availability zone",
-          "title": "Availability Zone",
-          "type": "string"
+          "title": "Availability Zone"
         },
         "backend": {
+          "const": "gcp",
           "default": "gcp",
           "description": "The volume backend",
-          "enum": [
-            "gcp"
-          ],
           "title": "Backend",
           "type": "string"
         },
         "name": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The volume name",
-          "title": "Name",
-          "type": "string"
+          "title": "Name"
         },
         "region": {
           "description": "The volume region",
@@ -1076,51 +1455,95 @@
           "type": "string"
         },
         "size": {
+          "anyOf": [
+            {
+              "type": "number"
+            },
+            {
+              "type": "integer"
+            },
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The volume size. Must be specified when creating new volumes",
-          "title": "Size",
-          "type": "number"
+          "title": "Size"
         },
         "tags": {
-          "additionalProperties": {
-            "type": "string"
-          },
+          "anyOf": [
+            {
+              "additionalProperties": {
+                "type": "string"
+              },
+              "type": "object"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The custom tags to associate with the volume. The tags are also propagated to the underlying backend resources. If there is a conflict with backend-level tags, does not override them",
-          "title": "Tags",
-          "type": "object"
+          "title": "Tags"
         },
         "type": {
+          "const": "volume",
           "default": "volume",
-          "enum": [
-            "volume"
-          ],
           "title": "Type",
           "type": "string"
         },
         "volume_id": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The volume ID. Must be specified when registering external volumes",
-          "title": "Volume Id",
-          "type": "string"
+          "title": "Volume Id"
         }
       },
       "required": [
         "region"
       ],
-      "title": "GCPVolumeConfigurationRequest",
+      "title": "GCPVolumeConfiguration",
       "type": "object"
     },
-    "GPUSpecRequest": {
+    "GPUSpec": {
       "additionalProperties": false,
       "properties": {
         "compute_capability": {
+          "anyOf": [
+            {
+              "type": "number"
+            },
+            {
+              "type": "string"
+            },
+            {
+              "items": {
+                "type": "integer"
+              },
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The minimum compute capability of the GPU (e.g., `7.5`)",
-          "items": {},
-          "title": "Compute Capability",
-          "type": "array"
+          "title": "Compute Capability"
         },
         "count": {
           "anyOf": [
             {
-              "$ref": "#/definitions/Range_int_"
+              "$ref": "#/$defs/Range_int_"
             },
             {
               "type": "integer"
@@ -1133,97 +1556,113 @@
             "max": null,
             "min": 1
           },
-          "description": "The number of GPUs",
-          "title": "Count"
+          "description": "The number of GPUs"
         },
         "memory": {
           "anyOf": [
             {
-              "$ref": "#/definitions/Range_Memory_"
+              "$ref": "#/$defs/Range_Memory_"
             },
             {
               "type": "integer"
             },
             {
               "type": "string"
+            },
+            {
+              "type": "null"
             }
           ],
-          "description": "The RAM size (e.g., `16GB`). Can be set to a range (e.g. `16GB..`, or `16GB..80GB`)",
-          "title": "Memory"
+          "default": null,
+          "description": "The RAM size (e.g., `16GB`). Can be set to a range (e.g. `16GB..`, or `16GB..80GB`)"
         },
         "name": {
           "anyOf": [
             {
+              "items": {
+                "type": "string"
+              },
               "type": "array"
             },
             {
               "type": "string"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The name of the GPU (e.g., `A100` or `H100`)",
-          "items": {
-            "type": "string"
-          },
           "title": "Name"
         },
         "total_memory": {
           "anyOf": [
             {
-              "$ref": "#/definitions/Range_Memory_"
+              "$ref": "#/$defs/Range_Memory_"
             },
             {
               "type": "integer"
             },
             {
               "type": "string"
+            },
+            {
+              "type": "null"
             }
           ],
-          "description": "The total RAM size (e.g., `32GB`). Can be set to a range (e.g. `16GB..`, or `16GB..80GB`)",
-          "title": "Total Memory"
+          "default": null,
+          "description": "The total RAM size (e.g., `32GB`). Can be set to a range (e.g. `16GB..`, or `16GB..80GB`)"
         },
         "vendor": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/AcceleratorVendor"
+            },
             {
-              "$ref": "#/definitions/AcceleratorVendor"
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The vendor of the GPU/accelerator, one of: `nvidia`, `amd`, `google` (alias: `tpu`), `intel`"
         }
       },
-      "title": "GPUSpecRequest",
+      "title": "GPUSpec",
       "type": "object"
     },
-    "GatewayConfigurationRequest": {
+    "GatewayConfiguration": {
       "additionalProperties": false,
       "properties": {
         "backend": {
-          "allOf": [
-            {
-              "$ref": "#/definitions/BackendType"
-            }
-          ],
+          "$ref": "#/$defs/BackendType",
           "description": "The gateway backend"
         },
         "certificate": {
-          "default": {
-            "type": "lets-encrypt"
-          },
-          "description": "The SSL certificate configuration. Set to `null` to disable. Defaults to `type: lets-encrypt`",
-          "discriminator": {
-            "mapping": {
-              "acm": "#/definitions/ACMGatewayCertificateRequest",
-              "lets-encrypt": "#/definitions/LetsEncryptGatewayCertificateRequest"
-            },
-            "propertyName": "type"
-          },
-          "oneOf": [
+          "anyOf": [
             {
-              "$ref": "#/definitions/LetsEncryptGatewayCertificateRequest"
+              "discriminator": {
+                "mapping": {
+                  "acm": "#/$defs/ACMGatewayCertificate",
+                  "lets-encrypt": "#/$defs/LetsEncryptGatewayCertificate"
+                },
+                "propertyName": "type"
+              },
+              "oneOf": [
+                {
+                  "$ref": "#/$defs/LetsEncryptGatewayCertificate"
+                },
+                {
+                  "$ref": "#/$defs/ACMGatewayCertificate"
+                }
+              ]
             },
             {
-              "$ref": "#/definitions/ACMGatewayCertificateRequest"
+              "type": "null"
             }
           ],
+          "default": {
+            "type": "lets-encrypt"
+          },
+          "description": "The SSL certificate configuration. Set to `null` to disable. Defaults to `type: lets-encrypt`",
           "title": "Certificate"
         },
         "default": {
@@ -1233,20 +1672,44 @@
           "type": "boolean"
         },
         "domain": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The gateway wildcard domain name, e.g. `example.com`. Service domain names are constructed as `<run name>.<gateway domain`. The domain name can use the `${{ run.project_name }}` variable to include the service\u2019s project name. Can be updated in-place. Updates do not affect existing services",
-          "title": "Domain",
-          "type": "string"
+          "title": "Domain"
         },
         "instance_type": {
+          "anyOf": [
+            {
+              "minLength": 1,
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "Backend-specific instance type to use for the gateway instance. Omit to use the backend's default, which is typically a small non-GPU instance",
-          "minLength": 1,
-          "title": "Instance Type",
-          "type": "string"
+          "title": "Instance Type"
         },
         "name": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The gateway name",
-          "title": "Name",
-          "type": "string"
+          "title": "Name"
         },
         "public_ip": {
           "default": true,
@@ -1260,33 +1723,50 @@
           "type": "string"
         },
         "replicas": {
+          "anyOf": [
+            {
+              "minimum": 1,
+              "type": "integer"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The number of gateway replicas. Defaults to `1`",
-          "minimum": 1,
-          "title": "Replicas",
-          "type": "integer"
+          "title": "Replicas"
         },
         "router": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/SGLangGatewayRouterConfig"
+            },
             {
-              "$ref": "#/definitions/SGLangGatewayRouterConfigRequest"
+              "type": "null"
             }
           ],
-          "description": "The router configuration for this gateway. E.g. `{ type: sglang, policy: round_robin }`.",
-          "title": "Router"
+          "default": null,
+          "description": "The router configuration for this gateway. E.g. `{ type: sglang, policy: round_robin }`."
         },
         "tags": {
-          "additionalProperties": {
-            "type": "string"
-          },
+          "anyOf": [
+            {
+              "additionalProperties": {
+                "type": "string"
+              },
+              "type": "object"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The custom tags to associate with the gateway. The tags are also propagated to the underlying backend resources. If there is a conflict with backend-level tags, does not override them",
-          "title": "Tags",
-          "type": "object"
+          "title": "Tags"
         },
         "type": {
+          "const": "gateway",
           "default": "gateway",
-          "enum": [
-            "gateway"
-          ],
           "title": "Type",
           "type": "string"
         }
@@ -1295,10 +1775,10 @@
         "backend",
         "region"
       ],
-      "title": "GatewayConfigurationRequest",
+      "title": "GatewayConfiguration",
       "type": "object"
     },
-    "HTTPHeaderSpecRequest": {
+    "HTTPHeaderSpec": {
       "additionalProperties": false,
       "properties": {
         "name": {
@@ -1320,10 +1800,10 @@
         "name",
         "value"
       ],
-      "title": "HTTPHeaderSpecRequest",
+      "title": "HTTPHeaderSpec",
       "type": "object"
     },
-    "HeaderPartitioningKeyRequest": {
+    "HeaderPartitioningKey": {
       "additionalProperties": false,
       "properties": {
         "header": {
@@ -1334,11 +1814,9 @@
           "type": "string"
         },
         "type": {
+          "const": "header",
           "default": "header",
           "description": "Partitioning type",
-          "enum": [
-            "header"
-          ],
           "title": "Type",
           "type": "string"
         }
@@ -1346,27 +1824,24 @@
       "required": [
         "header"
       ],
-      "title": "HeaderPartitioningKeyRequest",
+      "title": "HeaderPartitioningKey",
       "type": "object"
     },
-    "IPAddressPartitioningKeyRequest": {
+    "IPAddressPartitioningKey": {
       "additionalProperties": false,
       "properties": {
         "type": {
+          "const": "ip_address",
           "default": "ip_address",
           "description": "Partitioning type",
-          "enum": [
-            "ip_address"
-          ],
           "title": "Type",
           "type": "string"
         }
       },
-      "title": "IPAddressPartitioningKeyRequest",
+      "title": "IPAddressPartitioningKey",
       "type": "object"
     },
     "InstanceGroupPlacement": {
-      "description": "An enumeration.",
       "enum": [
         "any",
         "cluster"
@@ -1374,7 +1849,7 @@
       "title": "InstanceGroupPlacement",
       "type": "string"
     },
-    "InstanceHostnameSelectorRequest": {
+    "InstanceHostnameSelector": {
       "additionalProperties": false,
       "properties": {
         "hostname": {
@@ -1387,10 +1862,10 @@
       "required": [
         "hostname"
       ],
-      "title": "InstanceHostnameSelectorRequest",
+      "title": "InstanceHostnameSelector",
       "type": "object"
     },
-    "InstanceMountPointRequest": {
+    "InstanceMountPoint": {
       "additionalProperties": false,
       "properties": {
         "instance_path": {
@@ -1414,10 +1889,10 @@
         "instance_path",
         "path"
       ],
-      "title": "InstanceMountPointRequest",
+      "title": "InstanceMountPoint",
       "type": "object"
     },
-    "InstanceNameSelectorRequest": {
+    "InstanceNameSelector": {
       "additionalProperties": false,
       "properties": {
         "name": {
@@ -1430,10 +1905,10 @@
       "required": [
         "name"
       ],
-      "title": "InstanceNameSelectorRequest",
+      "title": "InstanceNameSelector",
       "type": "object"
     },
-    "KubernetesVolumeConfigurationRequest": {
+    "KubernetesVolumeConfiguration": {
       "additionalProperties": false,
       "properties": {
         "access_modes": {
@@ -1449,34 +1924,59 @@
         },
         "auto_cleanup_duration": {
           "anyOf": [
+            {
+              "const": "off",
+              "type": "string"
+            },
             {
               "type": "integer"
             },
             {
               "type": "string"
+            },
+            {
+              "type": "boolean"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "Time to wait after volume is no longer used by any job before deleting it. Defaults to keep the volume indefinitely. Use the value `off` or `-1` to disable auto-cleanup",
           "title": "Auto Cleanup Duration"
         },
         "backend": {
+          "const": "kubernetes",
           "default": "kubernetes",
           "description": "The volume backend",
-          "enum": [
-            "kubernetes"
-          ],
           "title": "Backend",
           "type": "string"
         },
         "claim_name": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The `PersistentVolumeClaim` name. Must be specified when registering the existing PVC instead of creating a new one",
-          "title": "Claim Name",
-          "type": "string"
+          "title": "Claim Name"
         },
         "name": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The volume name",
-          "title": "Name",
-          "type": "string"
+          "title": "Name"
         },
         "read_only": {
           "default": false,
@@ -1491,60 +1991,84 @@
           "type": "string"
         },
         "size": {
+          "anyOf": [
+            {
+              "type": "number"
+            },
+            {
+              "type": "integer"
+            },
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The requested volume size. Must be specified when creating new PVCs. Ignored if `claim_name` is set",
-          "title": "Size",
-          "type": "number"
+          "title": "Size"
         },
         "storage_class_name": {
-          "description": "The `StorageClass` name. Ignored if `claim_name` is set",
-          "title": "Storage Class Name",
-          "type": "string"
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
+          "description": "The `StorageClass` name. Ignored if `claim_name` is set",
+          "title": "Storage Class Name"
         },
         "tags": {
-          "additionalProperties": {
-            "type": "string"
-          },
+          "anyOf": [
+            {
+              "additionalProperties": {
+                "type": "string"
+              },
+              "type": "object"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The custom tags to associate with the volume. The tags are also propagated to the underlying backend resources. If there is a conflict with backend-level tags, does not override them",
-          "title": "Tags",
-          "type": "object"
+          "title": "Tags"
         },
         "type": {
+          "const": "volume",
           "default": "volume",
-          "enum": [
-            "volume"
-          ],
           "title": "Type",
           "type": "string"
         }
       },
-      "title": "KubernetesVolumeConfigurationRequest",
+      "title": "KubernetesVolumeConfiguration",
       "type": "object"
     },
-    "LetsEncryptGatewayCertificateRequest": {
+    "LetsEncryptGatewayCertificate": {
       "additionalProperties": false,
       "properties": {
         "type": {
+          "const": "lets-encrypt",
           "default": "lets-encrypt",
           "description": "Automatic certificates by Let's Encrypt",
-          "enum": [
-            "lets-encrypt"
-          ],
           "title": "Type",
           "type": "string"
         }
       },
-      "title": "LetsEncryptGatewayCertificateRequest",
+      "title": "LetsEncryptGatewayCertificate",
       "type": "object"
     },
-    "OpenAIChatModelRequest": {
+    "OpenAIChatModel": {
       "additionalProperties": false,
       "description": "Mapping of the model for the OpenAI-compatible endpoint.\n\nAttributes:\n    type (str): The type of the model, e.g. \"chat\"\n    name (str): The name of the model. This name will be used both to load model configuration from the HuggingFace Hub and in the OpenAI-compatible endpoint.\n    format (str): The format of the model, i.e. \"openai\".\n    prefix (str): The `base_url` prefix: `http://hostname/{prefix}/chat/completions`. Defaults to `/v1`.",
       "properties": {
         "format": {
+          "const": "openai",
           "description": "The serving format. Must be set to `openai`",
-          "enum": [
-            "openai"
-          ],
           "title": "Format",
           "type": "string"
         },
@@ -1560,11 +2084,9 @@
           "type": "string"
         },
         "type": {
+          "const": "chat",
           "default": "chat",
           "description": "The type of the model",
-          "enum": [
-            "chat"
-          ],
           "title": "Type",
           "type": "string"
         }
@@ -1573,10 +2095,10 @@
         "name",
         "format"
       ],
-      "title": "OpenAIChatModelRequest",
+      "title": "OpenAIChatModel",
       "type": "object"
     },
-    "PortMappingRequest": {
+    "PortMapping": {
       "additionalProperties": false,
       "properties": {
         "container_port": {
@@ -1586,33 +2108,49 @@
           "type": "integer"
         },
         "local_port": {
-          "exclusiveMinimum": 0,
-          "maximum": 65536,
-          "title": "Local Port",
-          "type": "integer"
+          "anyOf": [
+            {
+              "exclusiveMinimum": 0,
+              "maximum": 65536,
+              "type": "integer"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
+          "title": "Local Port"
         }
       },
       "required": [
         "container_port"
       ],
-      "title": "PortMappingRequest",
+      "title": "PortMapping",
       "type": "object"
     },
-    "ProbeConfigRequest": {
+    "ProbeConfig": {
       "additionalProperties": false,
       "properties": {
         "body": {
+          "anyOf": [
+            {
+              "maxLength": 2048,
+              "minLength": 1,
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The HTTP request body to send with the probe",
-          "maxLength": 2048,
-          "minLength": 1,
-          "title": "Body",
-          "type": "string"
+          "title": "Body"
         },
         "headers": {
           "default": [],
           "description": "A list of HTTP headers to include in the request",
           "items": {
-            "$ref": "#/definitions/HTTPHeaderSpecRequest"
+            "$ref": "#/$defs/HTTPHeaderSpec"
           },
           "maxItems": 16,
           "title": "Headers",
@@ -1625,29 +2163,49 @@
             },
             {
               "type": "string"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "Minimum amount of time between the end of one probe execution and the start of the next. Defaults to `15s`",
           "title": "Interval"
         },
         "method": {
+          "anyOf": [
+            {
+              "enum": [
+                "get",
+                "post",
+                "put",
+                "delete",
+                "patch",
+                "head"
+              ],
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The HTTP method to use for the probe (e.g., `get`, `post`, etc.). Defaults to `get`",
-          "enum": [
-            "get",
-            "post",
-            "put",
-            "delete",
-            "patch",
-            "head"
-          ],
-          "title": "Method",
-          "type": "string"
+          "title": "Method"
         },
         "ready_after": {
+          "anyOf": [
+            {
+              "minimum": 1,
+              "type": "integer"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The number of consecutive successful probe executions required for the replica to be considered ready. Used during rolling deployments. Defaults to `1`",
-          "minimum": 1,
-          "title": "Ready After",
-          "type": "integer"
+          "title": "Ready After"
         },
         "timeout": {
           "anyOf": [
@@ -1656,37 +2214,55 @@
             },
             {
               "type": "string"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "Maximum amount of time the HTTP request is allowed to take. Defaults to `10s`",
           "title": "Timeout"
         },
         "type": {
+          "const": "http",
           "description": "The probe type. Must be `http`",
-          "enum": [
-            "http"
-          ],
           "title": "Type",
           "type": "string"
         },
         "until_ready": {
+          "anyOf": [
+            {
+              "type": "boolean"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "If `true`, the probe will stop being executed as soon as it reaches the `ready_after` threshold of successful executions. Defaults to `false`",
-          "title": "Until Ready",
-          "type": "boolean"
+          "title": "Until Ready"
         },
         "url": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The URL to request. Defaults to `/`",
-          "title": "Url",
-          "type": "string"
+          "title": "Url"
         }
       },
       "required": [
         "type"
       ],
-      "title": "ProbeConfigRequest",
+      "title": "ProbeConfig",
       "type": "object"
     },
-    "ProfileRetryRequest": {
+    "ProfileRetry": {
       "additionalProperties": false,
       "properties": {
         "duration": {
@@ -1696,24 +2272,36 @@
             },
             {
               "type": "string"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The maximum period of retrying the run, e.g., `4h` or `1d`. The period is calculated as a run age for `no-capacity` event and as a time passed since the last `interruption` and `error` for `interruption` and `error` events.",
           "title": "Duration"
         },
         "on_events": {
+          "anyOf": [
+            {
+              "items": {
+                "$ref": "#/$defs/RetryEvent"
+              },
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The list of events that should be handled with retry. Supported events are `no-capacity`, `interruption`, `error`. Omit to retry on all events",
-          "items": {
-            "$ref": "#/definitions/RetryEvent"
-          },
-          "type": "array"
+          "title": "On Events"
         }
       },
-      "title": "ProfileRetryRequest",
+      "title": "ProfileRetry",
       "type": "object"
     },
     "PythonVersion": {
-      "description": "An enumeration.",
       "enum": [
         "3.9",
         "3.10",
@@ -1728,12 +2316,40 @@
       "additionalProperties": false,
       "properties": {
         "max": {
-          "title": "Max",
-          "type": "number"
+          "anyOf": [
+            {
+              "type": "number"
+            },
+            {
+              "type": "integer"
+            },
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
+          "title": "Max"
         },
         "min": {
-          "title": "Min",
-          "type": "number"
+          "anyOf": [
+            {
+              "type": "number"
+            },
+            {
+              "type": "integer"
+            },
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
+          "title": "Min"
         }
       },
       "title": "Range[Memory]",
@@ -1743,18 +2359,34 @@
       "additionalProperties": false,
       "properties": {
         "max": {
-          "title": "Max",
-          "type": "integer"
+          "anyOf": [
+            {
+              "type": "integer"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
+          "title": "Max"
         },
         "min": {
-          "title": "Min",
-          "type": "integer"
+          "anyOf": [
+            {
+              "type": "integer"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
+          "title": "Min"
         }
       },
       "title": "Range[int]",
       "type": "object"
     },
-    "RateLimitRequest": {
+    "RateLimit": {
       "additionalProperties": false,
       "properties": {
         "burst": {
@@ -1772,17 +2404,17 @@
           "description": "The partitioning key. Each incoming request belongs to a partition and rate limits are applied per partition. Defaults to partitioning by client IP address",
           "discriminator": {
             "mapping": {
-              "header": "#/definitions/HeaderPartitioningKeyRequest",
-              "ip_address": "#/definitions/IPAddressPartitioningKeyRequest"
+              "header": "#/$defs/HeaderPartitioningKey",
+              "ip_address": "#/$defs/IPAddressPartitioningKey"
             },
             "propertyName": "type"
           },
           "oneOf": [
             {
-              "$ref": "#/definitions/IPAddressPartitioningKeyRequest"
+              "$ref": "#/$defs/IPAddressPartitioningKey"
             },
             {
-              "$ref": "#/definitions/HeaderPartitioningKeyRequest"
+              "$ref": "#/$defs/HeaderPartitioningKey"
             }
           ],
           "title": "Key"
@@ -1806,10 +2438,10 @@
       "required": [
         "rps"
       ],
-      "title": "RateLimitRequest",
+      "title": "RateLimit",
       "type": "object"
     },
-    "RegistryAuthRequest": {
+    "RegistryAuth": {
       "additionalProperties": false,
       "description": "Credentials for pulling a private Docker image.\n\nAttributes:\n    username (str): The username\n    password (str): The password or access token",
       "properties": {
@@ -1828,10 +2460,10 @@
         "username",
         "password"
       ],
-      "title": "RegistryAuthRequest",
+      "title": "RegistryAuth",
       "type": "object"
     },
-    "ReplicaGroupRequest": {
+    "ReplicaGroup": {
       "additionalProperties": false,
       "properties": {
         "commands": {
@@ -1844,58 +2476,111 @@
           "type": "array"
         },
         "count": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/Range_int_"
+            },
+            {
+              "type": "integer"
+            },
             {
-              "$ref": "#/definitions/Range_int_"
+              "type": "string"
             }
           ],
-          "description": "The number of replicas. Can be a number (e.g. `2`) or a range (`0..4` or `1..8`). If it's a range, the `scaling` property is required",
-          "title": "Count"
+          "description": "The number of replicas. Can be a number (e.g. `2`) or a range (`0..4` or `1..8`). If it's a range, the `scaling` property is required"
         },
         "docker": {
+          "anyOf": [
+            {
+              "type": "boolean"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "Use the docker-in-docker image for this group (injects `start-dockerd` and runs privileged). Mutually exclusive with group-level `image`, `python`, and `nvcc`.",
-          "title": "Docker",
-          "type": "boolean"
+          "title": "Docker"
         },
         "image": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The name of the Docker image to run for replicas in this group. Mutually exclusive with group-level `docker` and `python`.",
-          "title": "Image",
-          "type": "string"
+          "title": "Image"
         },
         "name": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The name of the replica group. If not provided, defaults to '0', '1', etc. based on position.",
-          "title": "Name",
-          "type": "string"
+          "title": "Name"
         },
         "nvcc": {
+          "anyOf": [
+            {
+              "type": "boolean"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "Use the image with NVIDIA CUDA Compiler (NVCC) included for replicas in this group. Mutually exclusive with group-level `docker`.",
-          "title": "Nvcc",
-          "type": "boolean"
+          "title": "Nvcc"
         },
         "privileged": {
+          "anyOf": [
+            {
+              "type": "boolean"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "Run replicas in this group in privileged mode.",
-          "title": "Privileged",
-          "type": "boolean"
+          "title": "Privileged"
         },
         "python": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/PythonVersion"
+            },
             {
-              "$ref": "#/definitions/PythonVersion"
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The major version of Python for replicas in this group. Mutually exclusive with group-level `image` and `docker`."
         },
         "reservation": {
-          "description": "The existing reservation to use for replicas in this group. Supports AWS Capacity Reservations, AWS Capacity Blocks, and GCP reservations",
-          "title": "Reservation",
-          "type": "string"
-        },
-        "resources": {
-          "allOf": [
+          "anyOf": [
             {
-              "$ref": "#/definitions/ResourcesSpecRequest"
+              "type": "string"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
+          "description": "The existing reservation to use for replicas in this group. Supports AWS Capacity Reservations, AWS Capacity Blocks, and GCP reservations",
+          "title": "Reservation"
+        },
+        "resources": {
+          "$ref": "#/$defs/ResourcesSpec",
           "default": {
             "cpu": {
               "max": null,
@@ -1924,43 +2609,52 @@
             },
             "shm_size": null
           },
-          "description": "The resources requirements for replicas in this group",
-          "title": "Resources"
+          "description": "The resources requirements for replicas in this group"
         },
         "router": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/ReplicaGroupRouterConfig"
+            },
             {
-              "$ref": "#/definitions/ReplicaGroupRouterConfigRequest"
+              "type": "null"
             }
           ],
-          "description": "When set, replicas in this group run the in-service HTTP router (e.g. SGLang).",
-          "title": "Router"
+          "default": null,
+          "description": "When set, replicas in this group run the in-service HTTP router (e.g. SGLang)."
         },
         "scaling": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/ScalingSpec"
+            },
             {
-              "$ref": "#/definitions/ScalingSpecRequest"
+              "type": "null"
             }
           ],
-          "description": "The auto-scaling rules. Required if `count` is set to a range",
-          "title": "Scaling"
+          "default": null,
+          "description": "The auto-scaling rules. Required if `count` is set to a range"
         },
         "spot_policy": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/SpotPolicy"
+            },
             {
-              "$ref": "#/definitions/SpotPolicy"
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The policy for provisioning spot or on-demand instances for replicas in this group: `spot`, `on-demand`, `auto`"
         }
       },
       "required": [
         "count"
       ],
-      "title": "ReplicaGroupRequest",
+      "title": "ReplicaGroup",
       "type": "object"
     },
-    "ReplicaGroupRouterConfigRequest": {
+    "ReplicaGroupRouterConfig": {
       "additionalProperties": false,
       "properties": {
         "type": {
@@ -1974,11 +2668,10 @@
           "type": "string"
         }
       },
-      "title": "ReplicaGroupRouterConfigRequest",
+      "title": "ReplicaGroupRouterConfig",
       "type": "object"
     },
     "RepoExistsAction": {
-      "description": "An enumeration.",
       "enum": [
         "error",
         "skip"
@@ -1986,32 +2679,52 @@
       "title": "RepoExistsAction",
       "type": "string"
     },
-    "RepoSpecRequest": {
+    "RepoSpec": {
       "additionalProperties": false,
       "properties": {
         "branch": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The repo branch. Defaults to the active branch for local paths and the default branch for URLs",
-          "title": "Branch",
-          "type": "string"
+          "title": "Branch"
         },
         "hash": {
-          "description": "The commit hash",
-          "title": "Hash",
-          "type": "string"
-        },
-        "if_exists": {
-          "allOf": [
+          "anyOf": [
             {
-              "$ref": "#/definitions/RepoExistsAction"
+              "type": "string"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
+          "description": "The commit hash",
+          "title": "Hash"
+        },
+        "if_exists": {
+          "$ref": "#/$defs/RepoExistsAction",
           "default": "error",
           "description": "The action to be taken if `path` exists and is not empty. One of: `error`, `skip`"
         },
         "local_path": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The path to the Git repo on the user's machine. Relative paths are resolved relative to the parent directory of the the configuration file. Mutually exclusive with `url`",
-          "title": "Local Path",
-          "type": "string"
+          "title": "Local Path"
         },
         "path": {
           "default": ".",
@@ -2020,30 +2733,38 @@
           "type": "string"
         },
         "url": {
-          "description": "The Git repo URL. Mutually exclusive with `local_path`",
-          "title": "Url",
-          "type": "string"
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
+          "description": "The Git repo URL. Mutually exclusive with `local_path`",
+          "title": "Url"
         }
       },
-      "title": "RepoSpecRequest",
+      "title": "RepoSpec",
       "type": "object"
     },
-    "ResourcesSpecRequest": {
+    "ResourcesSpec": {
       "additionalProperties": false,
       "properties": {
         "cpu": {
           "anyOf": [
             {
-              "$ref": "#/definitions/CPUSpecRequest"
-            },
-            {
-              "$ref": "#/definitions/Range_int_"
+              "$ref": "#/$defs/CPUSpec"
             },
             {
               "type": "integer"
             },
             {
               "type": "string"
+            },
+            {
+              "$ref": "#/$defs/Range_int_"
             }
           ],
           "default": {
@@ -2059,13 +2780,16 @@
         "disk": {
           "anyOf": [
             {
-              "$ref": "#/definitions/DiskSpecRequest"
+              "$ref": "#/$defs/DiskSpec"
             },
             {
               "type": "integer"
             },
             {
               "type": "string"
+            },
+            {
+              "type": "null"
             }
           ],
           "default": {
@@ -2074,19 +2798,21 @@
               "min": 100.0
             }
           },
-          "description": "The disk resources",
-          "title": "Disk"
+          "description": "The disk resources"
         },
         "gpu": {
           "anyOf": [
             {
-              "$ref": "#/definitions/GPUSpecRequest"
+              "$ref": "#/$defs/GPUSpec"
             },
             {
               "type": "integer"
             },
             {
               "type": "string"
+            },
+            {
+              "type": "null"
             }
           ],
           "default": {
@@ -2100,13 +2826,12 @@
             "total_memory": null,
             "vendor": null
           },
-          "description": "The GPU requirements",
-          "title": "Gpu"
+          "description": "The GPU requirements"
         },
         "memory": {
           "anyOf": [
             {
-              "$ref": "#/definitions/Range_Memory_"
+              "$ref": "#/$defs/Range_Memory_"
             },
             {
               "type": "integer"
@@ -2119,8 +2844,7 @@
             "max": null,
             "min": 8.0
           },
-          "description": "The RAM size (e.g., `8GB`)",
-          "title": "Memory"
+          "description": "The RAM size (e.g., `8GB`)"
         },
         "shm_size": {
           "anyOf": [
@@ -2132,17 +2856,20 @@
             },
             {
               "type": "string"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The size of shared memory (e.g., `8GB`). If you are using parallel communicating processes (e.g., dataloaders in PyTorch), you may need to configure this",
           "title": "Shm Size"
         }
       },
-      "title": "ResourcesSpecRequest",
+      "title": "ResourcesSpec",
       "type": "object"
     },
     "RetryEvent": {
-      "description": "An enumeration.",
       "enum": [
         "no-capacity",
         "interruption",
@@ -2151,38 +2878,63 @@
       "title": "RetryEvent",
       "type": "string"
     },
-    "RunpodVolumeConfigurationRequest": {
+    "RunpodVolumeConfiguration": {
       "additionalProperties": false,
       "properties": {
         "auto_cleanup_duration": {
           "anyOf": [
+            {
+              "const": "off",
+              "type": "string"
+            },
             {
               "type": "integer"
             },
             {
               "type": "string"
+            },
+            {
+              "type": "boolean"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "Time to wait after volume is no longer used by any job before deleting it. Defaults to keep the volume indefinitely. Use the value `off` or `-1` to disable auto-cleanup",
           "title": "Auto Cleanup Duration"
         },
         "availability_zone": {
-          "title": "Availability Zone",
-          "type": "string"
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
+          "title": "Availability Zone"
         },
         "backend": {
+          "const": "runpod",
           "default": "runpod",
           "description": "The volume backend",
-          "enum": [
-            "runpod"
-          ],
           "title": "Backend",
           "type": "string"
         },
         "name": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The volume name",
-          "title": "Name",
-          "type": "string"
+          "title": "Name"
         },
         "region": {
           "description": "The volume region",
@@ -2190,39 +2942,67 @@
           "type": "string"
         },
         "size": {
+          "anyOf": [
+            {
+              "type": "number"
+            },
+            {
+              "type": "integer"
+            },
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The volume size. Must be specified when creating new volumes",
-          "title": "Size",
-          "type": "number"
+          "title": "Size"
         },
         "tags": {
-          "additionalProperties": {
-            "type": "string"
-          },
+          "anyOf": [
+            {
+              "additionalProperties": {
+                "type": "string"
+              },
+              "type": "object"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The custom tags to associate with the volume. The tags are also propagated to the underlying backend resources. If there is a conflict with backend-level tags, does not override them",
-          "title": "Tags",
-          "type": "object"
+          "title": "Tags"
         },
         "type": {
+          "const": "volume",
           "default": "volume",
-          "enum": [
-            "volume"
-          ],
           "title": "Type",
           "type": "string"
         },
         "volume_id": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The volume ID. Must be specified when registering external volumes",
-          "title": "Volume Id",
-          "type": "string"
+          "title": "Volume Id"
         }
       },
       "required": [
         "region"
       ],
-      "title": "RunpodVolumeConfigurationRequest",
+      "title": "RunpodVolumeConfiguration",
       "type": "object"
     },
-    "SGLangGatewayRouterConfigRequest": {
+    "SGLangGatewayRouterConfig": {
       "additionalProperties": false,
       "description": "Gateway-level router configuration. type and policy only. pd_disaggregation is service-level.",
       "properties": {
@@ -2239,19 +3019,17 @@
           "type": "string"
         },
         "type": {
+          "const": "sglang",
           "default": "sglang",
           "description": "The router type enabled on this gateway.",
-          "enum": [
-            "sglang"
-          ],
           "title": "Type",
           "type": "string"
         }
       },
-      "title": "SGLangGatewayRouterConfigRequest",
+      "title": "SGLangGatewayRouterConfig",
       "type": "object"
     },
-    "SGLangServiceRouterConfigRequest": {
+    "SGLangServiceRouterConfig": {
       "additionalProperties": false,
       "properties": {
         "pd_disaggregation": {
@@ -2273,34 +3051,34 @@
           "type": "string"
         },
         "type": {
+          "const": "sglang",
           "default": "sglang",
           "description": "The router type",
-          "enum": [
-            "sglang"
-          ],
           "title": "Type",
           "type": "string"
         }
       },
-      "title": "SGLangServiceRouterConfigRequest",
+      "title": "SGLangServiceRouterConfig",
       "type": "object"
     },
-    "SSHHostParamsRequest": {
+    "SSHHostParams": {
       "additionalProperties": false,
       "properties": {
         "blocks": {
           "anyOf": [
             {
-              "enum": [
-                "auto"
-              ],
+              "const": "auto",
               "type": "string"
             },
             {
               "minimum": 1,
               "type": "integer"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The amount of blocks to split the instance into, a number or `auto`. `auto` means as many as possible. The number of GPUs and CPUs must be divisible by the number of blocks. Defaults to the top-level `blocks` value",
           "title": "Blocks"
         },
@@ -2310,50 +3088,101 @@
           "type": "string"
         },
         "identity_file": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The private key to use for this host",
-          "title": "Identity File",
-          "type": "string"
+          "title": "Identity File"
         },
         "internal_ip": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The internal IP of the host used for communication inside the cluster. If not specified, `dstack` will use the IP address from `network` or from the first found internal network.",
-          "title": "Internal Ip",
-          "type": "string"
+          "title": "Internal Ip"
         },
         "port": {
+          "anyOf": [
+            {
+              "type": "integer"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The SSH port to connect to for this host",
-          "title": "Port",
-          "type": "integer"
+          "title": "Port"
         },
         "proxy_jump": {
-          "allOf": [
+          "anyOf": [
             {
-              "$ref": "#/definitions/SSHProxyParamsRequest"
+              "$ref": "#/$defs/SSHProxyParams"
+            },
+            {
+              "type": "null"
             }
           ],
-          "description": "The SSH proxy configuration for this host",
-          "title": "Proxy Jump"
+          "default": null,
+          "description": "The SSH proxy configuration for this host"
         },
         "ssh_key": {
-          "$ref": "#/definitions/SSHKeyRequest"
+          "anyOf": [
+            {
+              "$ref": "#/$defs/SSHKey"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null
         },
         "user": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The user to log in with for this host",
-          "title": "User",
-          "type": "string"
+          "title": "User"
         }
       },
       "required": [
         "hostname"
       ],
-      "title": "SSHHostParamsRequest",
+      "title": "SSHHostParams",
       "type": "object"
     },
-    "SSHKeyRequest": {
+    "SSHKey": {
       "additionalProperties": false,
       "properties": {
         "private": {
-          "title": "Private",
-          "type": "string"
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
+          "title": "Private"
         },
         "public": {
           "title": "Public",
@@ -2363,10 +3192,10 @@
       "required": [
         "public"
       ],
-      "title": "SSHKeyRequest",
+      "title": "SSHKey",
       "type": "object"
     },
-    "SSHParamsRequest": {
+    "SSHParams": {
       "additionalProperties": false,
       "properties": {
         "hosts": {
@@ -2374,7 +3203,7 @@
           "items": {
             "anyOf": [
               {
-                "$ref": "#/definitions/SSHHostParamsRequest"
+                "$ref": "#/$defs/SSHHostParams"
               },
               {
                 "type": "string"
@@ -2385,45 +3214,88 @@
           "type": "array"
         },
         "identity_file": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The private key to use for all hosts",
-          "title": "Identity File",
-          "type": "string"
+          "title": "Identity File"
         },
         "network": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The network address for cluster setup in the format `<ip>/<netmask>`. `dstack` will use IP addresses from this network for communication between hosts. If not specified, `dstack` will use IPs from the first found internal network.",
-          "title": "Network",
-          "type": "string"
+          "title": "Network"
         },
         "port": {
+          "anyOf": [
+            {
+              "type": "integer"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The SSH port to connect to",
-          "title": "Port",
-          "type": "integer"
+          "title": "Port"
         },
         "proxy_jump": {
-          "allOf": [
+          "anyOf": [
             {
-              "$ref": "#/definitions/SSHProxyParamsRequest"
+              "$ref": "#/$defs/SSHProxyParams"
+            },
+            {
+              "type": "null"
             }
           ],
-          "description": "The SSH proxy configuration for all hosts",
-          "title": "Proxy Jump"
+          "default": null,
+          "description": "The SSH proxy configuration for all hosts"
         },
         "ssh_key": {
-          "$ref": "#/definitions/SSHKeyRequest"
+          "anyOf": [
+            {
+              "$ref": "#/$defs/SSHKey"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null
         },
         "user": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The user to log in with on all hosts",
-          "title": "User",
-          "type": "string"
+          "title": "User"
         }
       },
       "required": [
         "hosts"
       ],
-      "title": "SSHParamsRequest",
+      "title": "SSHParams",
       "type": "object"
     },
-    "SSHProxyParamsRequest": {
+    "SSHProxyParams": {
       "additionalProperties": false,
       "properties": {
         "hostname": {
@@ -2437,12 +3309,28 @@
           "type": "string"
         },
         "port": {
+          "anyOf": [
+            {
+              "type": "integer"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The SSH port of proxy host",
-          "title": "Port",
-          "type": "integer"
+          "title": "Port"
         },
         "ssh_key": {
-          "$ref": "#/definitions/SSHKeyRequest"
+          "anyOf": [
+            {
+              "$ref": "#/$defs/SSHKey"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null
         },
         "user": {
           "description": "The user to log in with for proxy host",
@@ -2455,31 +3343,43 @@
         "user",
         "identity_file"
       ],
-      "title": "SSHProxyParamsRequest",
+      "title": "SSHProxyParams",
       "type": "object"
     },
-    "ScalingSpecRequest": {
+    "ScalingSpec": {
       "additionalProperties": false,
       "properties": {
         "metric": {
+          "const": "rps",
           "description": "The target metric to track. Currently, the only supported value is `rps` (meaning requests per second)",
-          "enum": [
-            "rps"
-          ],
           "title": "Metric",
           "type": "string"
         },
         "scale_down_delay": {
+          "anyOf": [
+            {
+              "type": "integer"
+            },
+            {
+              "type": "string"
+            }
+          ],
           "default": 600,
           "description": "The minimum time, in seconds, between a scaling event and the next scale-down decision. Used to prevent overly frequent scaling",
-          "title": "Scale Down Delay",
-          "type": "integer"
+          "title": "Scale Down Delay"
         },
         "scale_up_delay": {
+          "anyOf": [
+            {
+              "type": "integer"
+            },
+            {
+              "type": "string"
+            }
+          ],
           "default": 300,
           "description": "The minimum time, in seconds, between a scaling event and the next scale-up decision. Used to prevent overly frequent scaling",
-          "title": "Scale Up Delay",
-          "type": "integer"
+          "title": "Scale Up Delay"
         },
         "target": {
           "description": "The target value of the metric. The number of replicas is calculated based on this number and automatically adjusts (scales up or down) as this metric changes",
@@ -2488,19 +3388,30 @@
           "type": "number"
         },
         "window": {
+          "anyOf": [
+            {
+              "type": "integer"
+            },
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The time window used to calculate requests per second. Allowed values: `30s`, `60s`, `300s`. Defaults to `60s`",
-          "title": "Window",
-          "type": "integer"
+          "title": "Window"
         }
       },
       "required": [
         "metric",
         "target"
       ],
-      "title": "ScalingSpecRequest",
+      "title": "ScalingSpec",
       "type": "object"
     },
-    "ScheduleRequest": {
+    "Schedule": {
       "additionalProperties": false,
       "properties": {
         "cron": {
@@ -2522,10 +3433,10 @@
       "required": [
         "cron"
       ],
-      "title": "ScheduleRequest",
+      "title": "Schedule",
       "type": "object"
     },
-    "ServiceConfigurationRequest": {
+    "ServiceConfiguration": {
       "additionalProperties": false,
       "properties": {
         "auth": {
@@ -2535,27 +3446,52 @@
           "type": "boolean"
         },
         "availability_zones": {
+          "anyOf": [
+            {
+              "items": {
+                "type": "string"
+              },
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The availability zones to consider for provisioning (e.g., `[eu-west-1a, us-west4-a]`)",
-          "items": {
-            "type": "string"
-          },
-          "title": "Availability Zones",
-          "type": "array"
+          "title": "Availability Zones"
         },
         "backend_options": {
+          "anyOf": [
+            {
+              "items": {
+                "$ref": "#/$defs/VastAIProfileOptions"
+              },
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "Backend-specific options, applied only to offers from that backend",
-          "items": {
-            "$ref": "#/definitions/VastAIProfileOptions"
-          },
-          "title": "Backend Options",
-          "type": "array"
+          "title": "Backend Options"
         },
         "backends": {
+          "anyOf": [
+            {
+              "items": {
+                "$ref": "#/$defs/BackendType"
+              },
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The backends to consider for provisioning (e.g., `[aws, gcp]`)",
-          "items": {
-            "$ref": "#/definitions/BackendType"
-          },
-          "type": "array"
+          "title": "Backends"
         },
         "commands": {
           "default": [],
@@ -2567,17 +3503,29 @@
           "type": "array"
         },
         "creation_policy": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/CreationPolicy"
+            },
             {
-              "$ref": "#/definitions/CreationPolicy"
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The policy for using instances from fleets: `reuse`, `reuse-or-create`. Defaults to `reuse-or-create`"
         },
         "docker": {
+          "anyOf": [
+            {
+              "type": "boolean"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "Use Docker inside the container. Mutually exclusive with `image`, `python`, and `nvcc`. Overrides `privileged`",
-          "title": "Docker",
-          "type": "boolean"
+          "title": "Docker"
         },
         "dstack": {
           "default": false,
@@ -2586,21 +3534,21 @@
           "type": "boolean"
         },
         "entrypoint": {
-          "description": "The Docker entrypoint",
-          "title": "Entrypoint",
-          "type": "string"
-        },
-        "env": {
-          "allOf": [
+          "anyOf": [
             {
-              "$ref": "#/definitions/Env"
+              "type": "string"
+            },
+            {
+              "type": "null"
             }
           ],
-          "default": {
-            "__root__": {}
-          },
-          "description": "The mapping or the list of environment variables",
-          "title": "Env"
+          "default": null,
+          "description": "The Docker entrypoint",
+          "title": "Entrypoint"
+        },
+        "env": {
+          "$ref": "#/$defs/Env",
+          "description": "The mapping or the list of environment variables"
         },
         "files": {
           "default": [],
@@ -2608,7 +3556,7 @@
           "items": {
             "anyOf": [
               {
-                "$ref": "#/definitions/FilePathMappingRequest"
+                "$ref": "#/$defs/FilePathMapping"
               },
               {
                 "type": "string"
@@ -2619,19 +3567,27 @@
           "type": "array"
         },
         "fleets": {
-          "description": "The fleets considered for reuse. For fleets owned by the current project, specify fleet names. For imported fleets, specify `<project name>/<fleet name>`",
-          "items": {
-            "anyOf": [
-              {
-                "$ref": "#/definitions/EntityReferenceRequest"
+          "anyOf": [
+            {
+              "items": {
+                "anyOf": [
+                  {
+                    "$ref": "#/$defs/EntityReference"
+                  },
+                  {
+                    "type": "string"
+                  }
+                ]
               },
-              {
-                "type": "string"
-              }
-            ]
-          },
-          "title": "Fleets",
-          "type": "array"
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
+          "description": "The fleets considered for reuse. For fleets owned by the current project, specify fleet names. For imported fleets, specify `<project name>/<fleet name>`",
+          "title": "Fleets"
         },
         "gateway": {
           "anyOf": [
@@ -2639,12 +3595,16 @@
               "type": "boolean"
             },
             {
-              "$ref": "#/definitions/EntityReferenceRequest"
+              "$ref": "#/$defs/EntityReference"
             },
             {
               "type": "string"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The name of the gateway. Specify boolean `false` to run without a gateway. Specify boolean `true` to run with the default gateway. Omit to run with the default gateway if there is one, or without a gateway otherwise",
           "title": "Gateway"
         },
@@ -2659,125 +3619,187 @@
               "type": "boolean"
             },
             {
-              "enum": [
-                "auto"
-              ],
+              "const": "auto",
               "type": "string"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "Enable HTTPS if running with a gateway. Set to `auto` to determine automatically based on gateway configuration. Defaults to `true`",
           "title": "Https"
         },
         "idle_duration": {
           "anyOf": [
+            {
+              "const": "off",
+              "type": "string"
+            },
             {
               "type": "integer"
             },
             {
               "type": "string"
+            },
+            {
+              "type": "boolean"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "Time to wait before terminating idle instances. When the run reuses an existing fleet instance, the fleet's `idle_duration` applies. When the run provisions a new instance, the shorter of the fleet's and run's values is used. Defaults to `5m` for runs and `3d` for fleets. Use `off` for unlimited duration. Only applied for VM-based backends",
           "title": "Idle Duration"
         },
         "image": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The name of the Docker image to run. If no `image` is specified, `dstack` uses an Ubuntu 24.04-based Docker image that comes pre-configured with `uv`, `python`, `pip`, the CUDA 13.0 runtime, InfiniBand, NCCL, and NCCL tests. It may also include provider-specific components such as EFA support on AWS. For non-Nvidia accelerators or NVidia GPUs unsupported by CUDA 13.0 (e.g. V100, P100), specify a custom Docker image.",
-          "title": "Image",
-          "type": "string"
+          "title": "Image"
         },
         "instance_types": {
+          "anyOf": [
+            {
+              "items": {
+                "type": "string"
+              },
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The cloud-specific instance types to consider for provisioning (e.g., `[g6e.24xlarge, n1-standard-4]`)",
-          "items": {
-            "type": "string"
-          },
-          "title": "Instance Types",
-          "type": "array"
+          "title": "Instance Types"
         },
         "instances": {
-          "description": "The specific fleet instances to consider for reuse. Each value can be an instance name string, or an object with `name`, `hostname`, or `fleet` and `instance`. When set, the run is only placed on matching existing instances.",
-          "items": {
-            "anyOf": [
-              {
-                "$ref": "#/definitions/InstanceNameSelectorRequest"
-              },
-              {
-                "$ref": "#/definitions/InstanceHostnameSelectorRequest"
-              },
-              {
-                "$ref": "#/definitions/FleetInstanceSelectorRequest"
-              },
-              {
-                "minLength": 1,
-                "type": "string"
-              }
-            ]
-          },
-          "minItems": 1,
-          "title": "Instances",
-          "type": "array"
+          "anyOf": [
+            {
+              "items": {
+                "anyOf": [
+                  {
+                    "$ref": "#/$defs/InstanceNameSelector"
+                  },
+                  {
+                    "$ref": "#/$defs/InstanceHostnameSelector"
+                  },
+                  {
+                    "$ref": "#/$defs/FleetInstanceSelector"
+                  },
+                  {
+                    "type": "string"
+                  }
+                ]
+              },
+              "minItems": 1,
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
+          "description": "The specific fleet instances to consider for reuse. Each value can be an instance name string, or an object with `name`, `hostname`, or `fleet` and `instance`. When set, the run is only placed on matching existing instances.",
+          "title": "Instances"
         },
         "max_duration": {
           "anyOf": [
             {
-              "enum": [
-                "off"
-              ],
-              "type": "string"
+              "type": "integer"
             },
             {
-              "type": "integer"
+              "type": "string"
             },
             {
               "type": "boolean"
             },
             {
-              "type": "string"
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The maximum duration of a run (e.g., `2h`, `1d`, etc) in a running state, excluding provisioning and pulling. After it elapses, the run is automatically stopped. Use `off` for unlimited duration. Defaults to `off`",
           "title": "Max Duration"
         },
         "max_price": {
+          "anyOf": [
+            {
+              "exclusiveMinimum": 0.0,
+              "type": "number"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The maximum instance price per hour, in dollars",
-          "exclusiveMinimum": 0.0,
-          "title": "Max Price",
-          "type": "number"
+          "title": "Max Price"
         },
         "model": {
           "anyOf": [
             {
               "discriminator": {
                 "mapping": {
-                  "openai": "#/definitions/OpenAIChatModelRequest",
-                  "tgi": "#/definitions/TGIChatModelRequest"
+                  "openai": "#/$defs/OpenAIChatModel",
+                  "tgi": "#/$defs/TGIChatModel"
                 },
                 "propertyName": "format"
               },
               "oneOf": [
                 {
-                  "$ref": "#/definitions/TGIChatModelRequest"
+                  "$ref": "#/$defs/TGIChatModel"
                 },
                 {
-                  "$ref": "#/definitions/OpenAIChatModelRequest"
+                  "$ref": "#/$defs/OpenAIChatModel"
                 }
               ]
             },
             {
               "type": "string"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "Mapping of the model for the OpenAI-compatible endpoint provided by `dstack`. Can be a full model format definition or just a model name. If it's a name, the service is expected to expose an OpenAI-compatible API at the `/v1` path",
           "title": "Model"
         },
         "name": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The run name. If not specified, a random name is generated",
-          "title": "Name",
-          "type": "string"
+          "title": "Name"
         },
         "nvcc": {
+          "anyOf": [
+            {
+              "type": "boolean"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "Use image with NVIDIA CUDA Compiler (NVCC) included. Mutually exclusive with `image` and `docker`",
-          "title": "Nvcc",
-          "type": "boolean"
+          "title": "Nvcc"
         },
         "port": {
           "anyOf": [
@@ -2791,18 +3813,26 @@
               "type": "string"
             },
             {
-              "$ref": "#/definitions/PortMappingRequest"
+              "$ref": "#/$defs/PortMapping"
             }
           ],
           "description": "The port the application listens on",
           "title": "Port"
         },
         "priority": {
+          "anyOf": [
+            {
+              "maximum": 100,
+              "minimum": 0,
+              "type": "integer"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The priority of the run, an integer between `0` and `100`. `dstack` tries to provision runs with higher priority first. Defaults to `0`",
-          "maximum": 100,
-          "minimum": 0,
-          "title": "Priority",
-          "type": "integer"
+          "title": "Priority"
         },
         "privileged": {
           "default": false,
@@ -2811,65 +3841,92 @@
           "type": "boolean"
         },
         "probes": {
+          "anyOf": [
+            {
+              "items": {
+                "$ref": "#/$defs/ProbeConfig"
+              },
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The list of probes to determine service health. If `model` is set, defaults to a `/v1/chat/completions` probe. Set explicitly to override",
-          "items": {
-            "$ref": "#/definitions/ProbeConfigRequest"
-          },
-          "title": "Probes",
-          "type": "array"
+          "title": "Probes"
         },
         "python": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/PythonVersion"
+            },
             {
-              "$ref": "#/definitions/PythonVersion"
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The major version of Python. Mutually exclusive with `image` and `docker`"
         },
         "rate_limits": {
           "default": [],
           "description": "Rate limiting rules",
           "items": {
-            "$ref": "#/definitions/RateLimitRequest"
+            "$ref": "#/$defs/RateLimit"
           },
           "title": "Rate Limits",
           "type": "array"
         },
         "regions": {
+          "anyOf": [
+            {
+              "items": {
+                "type": "string"
+              },
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The regions to consider for provisioning (e.g., `[eu-west-1, us-west4, westeurope]`)",
-          "items": {
-            "type": "string"
-          },
-          "title": "Regions",
-          "type": "array"
+          "title": "Regions"
         },
         "registry_auth": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/RegistryAuth"
+            },
             {
-              "$ref": "#/definitions/RegistryAuthRequest"
+              "type": "null"
             }
           ],
-          "description": "Credentials for pulling a private Docker image",
-          "title": "Registry Auth"
+          "default": null,
+          "description": "Credentials for pulling a private Docker image"
         },
         "replicas": {
           "anyOf": [
             {
               "items": {
-                "$ref": "#/definitions/ReplicaGroupRequest"
+                "$ref": "#/$defs/ReplicaGroup"
               },
               "type": "array"
             },
             {
-              "$ref": "#/definitions/Range_int_"
+              "$ref": "#/$defs/Range_int_"
             },
             {
               "type": "integer"
             },
             {
               "type": "string"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The number of replicas or a list of replica groups. Can be an integer (e.g., `2`), a range (e.g., `0..4`), or a list of replica groups. Each replica group defines replicas with shared configuration (commands, resources, scaling). When `replicas` is a list of replica groups, top-level `scaling`, `commands`, and `resources` are not allowed and must be specified in each replica group instead. ",
           "title": "Replicas"
         },
@@ -2877,22 +3934,33 @@
           "default": [],
           "description": "The list of Git repos",
           "items": {
-            "$ref": "#/definitions/RepoSpecRequest"
+            "anyOf": [
+              {
+                "$ref": "#/$defs/RepoSpec"
+              },
+              {
+                "type": "string"
+              }
+            ]
           },
           "title": "Repos",
           "type": "array"
         },
         "reservation": {
-          "description": "The existing reservation to use for instance provisioning. Supports AWS Capacity Reservations, AWS Capacity Blocks, and GCP reservations",
-          "title": "Reservation",
-          "type": "string"
-        },
-        "resources": {
-          "allOf": [
+          "anyOf": [
+            {
+              "type": "string"
+            },
             {
-              "$ref": "#/definitions/ResourcesSpecRequest"
+              "type": "null"
             }
           ],
+          "default": null,
+          "description": "The existing reservation to use for instance provisioning. Supports AWS Capacity Reservations, AWS Capacity Blocks, and GCP reservations",
+          "title": "Reservation"
+        },
+        "resources": {
+          "$ref": "#/$defs/ResourcesSpec",
           "default": {
             "cpu": {
               "max": null,
@@ -2921,47 +3989,59 @@
             },
             "shm_size": null
           },
-          "description": "The resources requirements to run the configuration",
-          "title": "Resources"
+          "description": "The resources requirements to run the configuration"
         },
         "retry": {
           "anyOf": [
             {
-              "$ref": "#/definitions/ProfileRetryRequest"
+              "$ref": "#/$defs/ProfileRetry"
             },
             {
               "type": "boolean"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The policy for resubmitting the run. Defaults to `false`",
           "title": "Retry"
         },
         "router": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/SGLangServiceRouterConfig"
+            },
             {
-              "$ref": "#/definitions/SGLangServiceRouterConfigRequest"
+              "type": "null"
             }
           ],
-          "description": "Router configuration for the service. Requires a gateway with matching router enabled. ",
-          "title": "Router"
+          "default": null,
+          "description": "Router configuration for the service. Requires a gateway with matching router enabled. "
         },
         "scaling": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/ScalingSpec"
+            },
             {
-              "$ref": "#/definitions/ScalingSpecRequest"
+              "type": "null"
             }
           ],
-          "description": "The auto-scaling rules. Required if `replicas` is set to a range",
-          "title": "Scaling"
+          "default": null,
+          "description": "The auto-scaling rules. Required if `replicas` is set to a range"
         },
         "schedule": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/Schedule"
+            },
             {
-              "$ref": "#/definitions/ScheduleRequest"
+              "type": "null"
             }
           ],
-          "description": "The schedule for starting the run at specified time",
-          "title": "Schedule"
+          "default": null,
+          "description": "The schedule for starting the run at specified time"
         },
         "setup": {
           "default": [],
@@ -2972,57 +4052,83 @@
           "type": "array"
         },
         "shell": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The shell used to run commands. Allowed values are `sh`, `bash`, or an absolute path, e.g., `/usr/bin/zsh`. Defaults to `/bin/sh` if the `image` is specified, `/bin/bash` otherwise",
-          "title": "Shell",
-          "type": "string"
+          "title": "Shell"
         },
         "single_branch": {
+          "anyOf": [
+            {
+              "type": "boolean"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "Whether to clone and track only the current branch or all remote branches. Relevant only when using remote Git repos. Defaults to `false` for dev environments and to `true` for tasks and services",
-          "title": "Single Branch",
-          "type": "boolean"
+          "title": "Single Branch"
         },
         "spot_policy": {
-          "allOf": [
+          "anyOf": [
             {
-              "$ref": "#/definitions/SpotPolicy"
+              "$ref": "#/$defs/SpotPolicy"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The policy for provisioning spot or on-demand instances: `spot`, `on-demand`, `auto`. Defaults to `on-demand`"
         },
         "startup_order": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/StartupOrder"
+            },
             {
-              "$ref": "#/definitions/StartupOrder"
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The order in which master and workers jobs are started: `any`, `master-first`, `workers-first`. Defaults to `any`"
         },
         "stop_criteria": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/StopCriteria"
+            },
             {
-              "$ref": "#/definitions/StopCriteria"
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The criteria determining when a multi-node run should be considered finished: `all-done`, `master-done`. Defaults to `all-done`"
         },
         "stop_duration": {
           "anyOf": [
             {
-              "enum": [
-                "off"
-              ],
-              "type": "string"
+              "type": "integer"
             },
             {
-              "type": "integer"
+              "type": "string"
             },
             {
               "type": "boolean"
             },
             {
-              "type": "string"
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The maximum duration of a run graceful stopping. After it elapses, the run is automatically forced stopped. This includes force detaching volumes used by the run. Use `off` for unlimited duration. Defaults to `5m`",
           "title": "Stop Duration"
         },
@@ -3033,34 +4139,51 @@
           "type": "boolean"
         },
         "tags": {
-          "additionalProperties": {
-            "type": "string"
-          },
+          "anyOf": [
+            {
+              "additionalProperties": {
+                "type": "string"
+              },
+              "type": "object"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The custom tags to associate with the resource. The tags are also propagated to the underlying backend resources. If there is a conflict with backend-level tags, does not override them",
-          "title": "Tags",
-          "type": "object"
+          "title": "Tags"
         },
         "type": {
+          "const": "service",
           "default": "service",
-          "enum": [
-            "service"
-          ],
           "title": "Type",
           "type": "string"
         },
         "user": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The user inside the container, `user_name_or_id[:group_name_or_id]` (e.g., `ubuntu`, `1000:1000`). Defaults to the default user from the `image`",
-          "title": "User",
-          "type": "string"
+          "title": "User"
         },
         "utilization_policy": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/UtilizationPolicy"
+            },
             {
-              "$ref": "#/definitions/UtilizationPolicyRequest"
+              "type": "null"
             }
           ],
-          "description": "Run termination policy based on utilization",
-          "title": "Utilization Policy"
+          "default": null,
+          "description": "Run termination policy based on utilization"
         },
         "volumes": {
           "default": [],
@@ -3068,10 +4191,10 @@
           "items": {
             "anyOf": [
               {
-                "$ref": "#/definitions/VolumeMountPointRequest"
+                "$ref": "#/$defs/VolumeMountPoint"
               },
               {
-                "$ref": "#/definitions/InstanceMountPointRequest"
+                "$ref": "#/$defs/InstanceMountPoint"
               },
               {
                 "type": "string"
@@ -3082,19 +4205,26 @@
           "type": "array"
         },
         "working_dir": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The absolute path to the working directory inside the container. Defaults to the `image`'s default working directory",
-          "title": "Working Dir",
-          "type": "string"
+          "title": "Working Dir"
         }
       },
       "required": [
         "port"
       ],
-      "title": "ServiceConfigurationRequest",
+      "title": "ServiceConfiguration",
       "type": "object"
     },
     "SpotPolicy": {
-      "description": "An enumeration.",
       "enum": [
         "spot",
         "on-demand",
@@ -3104,7 +4234,6 @@
       "type": "string"
     },
     "StartupOrder": {
-      "description": "An enumeration.",
       "enum": [
         "any",
         "master-first",
@@ -3114,7 +4243,6 @@
       "type": "string"
     },
     "StopCriteria": {
-      "description": "An enumeration.",
       "enum": [
         "all-done",
         "master-done"
@@ -3122,25 +4250,39 @@
       "title": "StopCriteria",
       "type": "string"
     },
-    "TGIChatModelRequest": {
+    "TGIChatModel": {
       "additionalProperties": false,
       "description": "Mapping of the model for the OpenAI-compatible endpoint.\n\nAttributes:\n    type (str): The type of the model, e.g. \"chat\"\n    name (str): The name of the model. This name will be used both to load model configuration from the HuggingFace Hub and in the OpenAI-compatible endpoint.\n    format (str): The format of the model, e.g. \"tgi\" if the model is served with HuggingFace's Text Generation Inference.\n    chat_template (Optional[str]): The custom prompt template for the model. If not specified, the default prompt template from the HuggingFace Hub configuration will be used.\n    eos_token (Optional[str]): The custom end of sentence token. If not specified, the default end of sentence token from the HuggingFace Hub configuration will be used.",
       "properties": {
         "chat_template": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The custom prompt template for the model. If not specified, the default prompt template from the HuggingFace Hub configuration will be used",
-          "title": "Chat Template",
-          "type": "string"
+          "title": "Chat Template"
         },
         "eos_token": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The custom end of sentence token. If not specified, the default end of sentence token from the HuggingFace Hub configuration will be used",
-          "title": "Eos Token",
-          "type": "string"
+          "title": "Eos Token"
         },
         "format": {
+          "const": "tgi",
           "description": "The serving format. Must be set to `tgi`",
-          "enum": [
-            "tgi"
-          ],
           "title": "Format",
           "type": "string"
         },
@@ -3150,11 +4292,9 @@
           "type": "string"
         },
         "type": {
+          "const": "chat",
           "default": "chat",
           "description": "The type of the model",
-          "enum": [
-            "chat"
-          ],
           "title": "Type",
           "type": "string"
         }
@@ -3163,34 +4303,59 @@
         "name",
         "format"
       ],
-      "title": "TGIChatModelRequest",
+      "title": "TGIChatModel",
       "type": "object"
     },
-    "TaskConfigurationRequest": {
+    "TaskConfiguration": {
       "additionalProperties": false,
       "properties": {
         "availability_zones": {
+          "anyOf": [
+            {
+              "items": {
+                "type": "string"
+              },
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The availability zones to consider for provisioning (e.g., `[eu-west-1a, us-west4-a]`)",
-          "items": {
-            "type": "string"
-          },
-          "title": "Availability Zones",
-          "type": "array"
+          "title": "Availability Zones"
         },
         "backend_options": {
+          "anyOf": [
+            {
+              "items": {
+                "$ref": "#/$defs/VastAIProfileOptions"
+              },
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "Backend-specific options, applied only to offers from that backend",
-          "items": {
-            "$ref": "#/definitions/VastAIProfileOptions"
-          },
-          "title": "Backend Options",
-          "type": "array"
+          "title": "Backend Options"
         },
         "backends": {
+          "anyOf": [
+            {
+              "items": {
+                "$ref": "#/$defs/BackendType"
+              },
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The backends to consider for provisioning (e.g., `[aws, gcp]`)",
-          "items": {
-            "$ref": "#/definitions/BackendType"
-          },
-          "type": "array"
+          "title": "Backends"
         },
         "commands": {
           "default": [],
@@ -3202,17 +4367,29 @@
           "type": "array"
         },
         "creation_policy": {
-          "allOf": [
+          "anyOf": [
             {
-              "$ref": "#/definitions/CreationPolicy"
+              "$ref": "#/$defs/CreationPolicy"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The policy for using instances from fleets: `reuse`, `reuse-or-create`. Defaults to `reuse-or-create`"
         },
         "docker": {
+          "anyOf": [
+            {
+              "type": "boolean"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "Use Docker inside the container. Mutually exclusive with `image`, `python`, and `nvcc`. Overrides `privileged`",
-          "title": "Docker",
-          "type": "boolean"
+          "title": "Docker"
         },
         "dstack": {
           "default": false,
@@ -3221,21 +4398,21 @@
           "type": "boolean"
         },
         "entrypoint": {
-          "description": "The Docker entrypoint",
-          "title": "Entrypoint",
-          "type": "string"
-        },
-        "env": {
-          "allOf": [
+          "anyOf": [
+            {
+              "type": "string"
+            },
             {
-              "$ref": "#/definitions/Env"
+              "type": "null"
             }
           ],
-          "default": {
-            "__root__": {}
-          },
-          "description": "The mapping or the list of environment variables",
-          "title": "Env"
+          "default": null,
+          "description": "The Docker entrypoint",
+          "title": "Entrypoint"
+        },
+        "env": {
+          "$ref": "#/$defs/Env",
+          "description": "The mapping or the list of environment variables"
         },
         "files": {
           "default": [],
@@ -3243,7 +4420,7 @@
           "items": {
             "anyOf": [
               {
-                "$ref": "#/definitions/FilePathMappingRequest"
+                "$ref": "#/$defs/FilePathMapping"
               },
               {
                 "type": "string"
@@ -3254,19 +4431,27 @@
           "type": "array"
         },
         "fleets": {
-          "description": "The fleets considered for reuse. For fleets owned by the current project, specify fleet names. For imported fleets, specify `<project name>/<fleet name>`",
-          "items": {
-            "anyOf": [
-              {
-                "$ref": "#/definitions/EntityReferenceRequest"
+          "anyOf": [
+            {
+              "items": {
+                "anyOf": [
+                  {
+                    "$ref": "#/$defs/EntityReference"
+                  },
+                  {
+                    "type": "string"
+                  }
+                ]
               },
-              {
-                "type": "string"
-              }
-            ]
-          },
-          "title": "Fleets",
-          "type": "array"
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
+          "description": "The fleets considered for reuse. For fleets owned by the current project, specify fleet names. For imported fleets, specify `<project name>/<fleet name>`",
+          "title": "Fleets"
         },
         "home_dir": {
           "default": "/root",
@@ -3275,83 +4460,131 @@
         },
         "idle_duration": {
           "anyOf": [
+            {
+              "const": "off",
+              "type": "string"
+            },
             {
               "type": "integer"
             },
             {
               "type": "string"
+            },
+            {
+              "type": "boolean"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "Time to wait before terminating idle instances. When the run reuses an existing fleet instance, the fleet's `idle_duration` applies. When the run provisions a new instance, the shorter of the fleet's and run's values is used. Defaults to `5m` for runs and `3d` for fleets. Use `off` for unlimited duration. Only applied for VM-based backends",
           "title": "Idle Duration"
         },
         "image": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The name of the Docker image to run. If no `image` is specified, `dstack` uses an Ubuntu 24.04-based Docker image that comes pre-configured with `uv`, `python`, `pip`, the CUDA 13.0 runtime, InfiniBand, NCCL, and NCCL tests. It may also include provider-specific components such as EFA support on AWS. For non-Nvidia accelerators or NVidia GPUs unsupported by CUDA 13.0 (e.g. V100, P100), specify a custom Docker image.",
-          "title": "Image",
-          "type": "string"
+          "title": "Image"
         },
         "instance_types": {
+          "anyOf": [
+            {
+              "items": {
+                "type": "string"
+              },
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The cloud-specific instance types to consider for provisioning (e.g., `[g6e.24xlarge, n1-standard-4]`)",
-          "items": {
-            "type": "string"
-          },
-          "title": "Instance Types",
-          "type": "array"
+          "title": "Instance Types"
         },
         "instances": {
-          "description": "The specific fleet instances to consider for reuse. Each value can be an instance name string, or an object with `name`, `hostname`, or `fleet` and `instance`. When set, the run is only placed on matching existing instances.",
-          "items": {
-            "anyOf": [
-              {
-                "$ref": "#/definitions/InstanceNameSelectorRequest"
-              },
-              {
-                "$ref": "#/definitions/InstanceHostnameSelectorRequest"
-              },
-              {
-                "$ref": "#/definitions/FleetInstanceSelectorRequest"
+          "anyOf": [
+            {
+              "items": {
+                "anyOf": [
+                  {
+                    "$ref": "#/$defs/InstanceNameSelector"
+                  },
+                  {
+                    "$ref": "#/$defs/InstanceHostnameSelector"
+                  },
+                  {
+                    "$ref": "#/$defs/FleetInstanceSelector"
+                  },
+                  {
+                    "type": "string"
+                  }
+                ]
               },
-              {
-                "minLength": 1,
-                "type": "string"
-              }
-            ]
-          },
-          "minItems": 1,
-          "title": "Instances",
-          "type": "array"
+              "minItems": 1,
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
+          "description": "The specific fleet instances to consider for reuse. Each value can be an instance name string, or an object with `name`, `hostname`, or `fleet` and `instance`. When set, the run is only placed on matching existing instances.",
+          "title": "Instances"
         },
         "max_duration": {
           "anyOf": [
             {
-              "enum": [
-                "off"
-              ],
-              "type": "string"
+              "type": "integer"
             },
             {
-              "type": "integer"
+              "type": "string"
             },
             {
               "type": "boolean"
             },
             {
-              "type": "string"
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The maximum duration of a run (e.g., `2h`, `1d`, etc) in a running state, excluding provisioning and pulling. After it elapses, the run is automatically stopped. Use `off` for unlimited duration. Defaults to `off`",
           "title": "Max Duration"
         },
         "max_price": {
+          "anyOf": [
+            {
+              "exclusiveMinimum": 0.0,
+              "type": "number"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The maximum instance price per hour, in dollars",
-          "exclusiveMinimum": 0.0,
-          "title": "Max Price",
-          "type": "number"
+          "title": "Max Price"
         },
         "name": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The run name. If not specified, a random name is generated",
-          "title": "Name",
-          "type": "string"
+          "title": "Name"
         },
         "nodes": {
           "default": 1,
@@ -3361,9 +4594,17 @@
           "type": "integer"
         },
         "nvcc": {
+          "anyOf": [
+            {
+              "type": "boolean"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "Use image with NVIDIA CUDA Compiler (NVCC) included. Mutually exclusive with `image` and `docker`",
-          "title": "Nvcc",
-          "type": "boolean"
+          "title": "Nvcc"
         },
         "ports": {
           "default": [],
@@ -3380,7 +4621,7 @@
                 "type": "string"
               },
               {
-                "$ref": "#/definitions/PortMappingRequest"
+                "$ref": "#/$defs/PortMapping"
               }
             ]
           },
@@ -3388,11 +4629,19 @@
           "type": "array"
         },
         "priority": {
+          "anyOf": [
+            {
+              "maximum": 100,
+              "minimum": 0,
+              "type": "integer"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The priority of the run, an integer between `0` and `100`. `dstack` tries to provision runs with higher priority first. Defaults to `0`",
-          "maximum": 100,
-          "minimum": 0,
-          "title": "Priority",
-          "type": "integer"
+          "title": "Priority"
         },
         "privileged": {
           "default": false,
@@ -3401,50 +4650,76 @@
           "type": "boolean"
         },
         "python": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/PythonVersion"
+            },
             {
-              "$ref": "#/definitions/PythonVersion"
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The major version of Python. Mutually exclusive with `image` and `docker`"
         },
         "regions": {
+          "anyOf": [
+            {
+              "items": {
+                "type": "string"
+              },
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The regions to consider for provisioning (e.g., `[eu-west-1, us-west4, westeurope]`)",
-          "items": {
-            "type": "string"
-          },
-          "title": "Regions",
-          "type": "array"
+          "title": "Regions"
         },
         "registry_auth": {
-          "allOf": [
+          "anyOf": [
             {
-              "$ref": "#/definitions/RegistryAuthRequest"
+              "$ref": "#/$defs/RegistryAuth"
+            },
+            {
+              "type": "null"
             }
           ],
-          "description": "Credentials for pulling a private Docker image",
-          "title": "Registry Auth"
+          "default": null,
+          "description": "Credentials for pulling a private Docker image"
         },
         "repos": {
           "default": [],
           "description": "The list of Git repos",
           "items": {
-            "$ref": "#/definitions/RepoSpecRequest"
+            "anyOf": [
+              {
+                "$ref": "#/$defs/RepoSpec"
+              },
+              {
+                "type": "string"
+              }
+            ]
           },
           "title": "Repos",
           "type": "array"
         },
         "reservation": {
-          "description": "The existing reservation to use for instance provisioning. Supports AWS Capacity Reservations, AWS Capacity Blocks, and GCP reservations",
-          "title": "Reservation",
-          "type": "string"
-        },
-        "resources": {
-          "allOf": [
+          "anyOf": [
+            {
+              "type": "string"
+            },
             {
-              "$ref": "#/definitions/ResourcesSpecRequest"
+              "type": "null"
             }
           ],
+          "default": null,
+          "description": "The existing reservation to use for instance provisioning. Supports AWS Capacity Reservations, AWS Capacity Blocks, and GCP reservations",
+          "title": "Reservation"
+        },
+        "resources": {
+          "$ref": "#/$defs/ResourcesSpec",
           "default": {
             "cpu": {
               "max": null,
@@ -3473,29 +4748,35 @@
             },
             "shm_size": null
           },
-          "description": "The resources requirements to run the configuration",
-          "title": "Resources"
+          "description": "The resources requirements to run the configuration"
         },
         "retry": {
           "anyOf": [
             {
-              "$ref": "#/definitions/ProfileRetryRequest"
+              "$ref": "#/$defs/ProfileRetry"
             },
             {
               "type": "boolean"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The policy for resubmitting the run. Defaults to `false`",
           "title": "Retry"
         },
         "schedule": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/Schedule"
+            },
             {
-              "$ref": "#/definitions/ScheduleRequest"
+              "type": "null"
             }
           ],
-          "description": "The schedule for starting the run at specified time",
-          "title": "Schedule"
+          "default": null,
+          "description": "The schedule for starting the run at specified time"
         },
         "setup": {
           "default": [],
@@ -3506,89 +4787,132 @@
           "type": "array"
         },
         "shell": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The shell used to run commands. Allowed values are `sh`, `bash`, or an absolute path, e.g., `/usr/bin/zsh`. Defaults to `/bin/sh` if the `image` is specified, `/bin/bash` otherwise",
-          "title": "Shell",
-          "type": "string"
+          "title": "Shell"
         },
         "single_branch": {
+          "anyOf": [
+            {
+              "type": "boolean"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "Whether to clone and track only the current branch or all remote branches. Relevant only when using remote Git repos. Defaults to `false` for dev environments and to `true` for tasks and services",
-          "title": "Single Branch",
-          "type": "boolean"
+          "title": "Single Branch"
         },
         "spot_policy": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/SpotPolicy"
+            },
             {
-              "$ref": "#/definitions/SpotPolicy"
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The policy for provisioning spot or on-demand instances: `spot`, `on-demand`, `auto`. Defaults to `on-demand`"
         },
         "startup_order": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/StartupOrder"
+            },
             {
-              "$ref": "#/definitions/StartupOrder"
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The order in which master and workers jobs are started: `any`, `master-first`, `workers-first`. Defaults to `any`"
         },
         "stop_criteria": {
-          "allOf": [
+          "anyOf": [
             {
-              "$ref": "#/definitions/StopCriteria"
+              "$ref": "#/$defs/StopCriteria"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The criteria determining when a multi-node run should be considered finished: `all-done`, `master-done`. Defaults to `all-done`"
         },
         "stop_duration": {
           "anyOf": [
             {
-              "enum": [
-                "off"
-              ],
-              "type": "string"
+              "type": "integer"
             },
             {
-              "type": "integer"
+              "type": "string"
             },
             {
               "type": "boolean"
             },
             {
-              "type": "string"
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The maximum duration of a run graceful stopping. After it elapses, the run is automatically forced stopped. This includes force detaching volumes used by the run. Use `off` for unlimited duration. Defaults to `5m`",
           "title": "Stop Duration"
         },
         "tags": {
-          "additionalProperties": {
-            "type": "string"
-          },
+          "anyOf": [
+            {
+              "additionalProperties": {
+                "type": "string"
+              },
+              "type": "object"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The custom tags to associate with the resource. The tags are also propagated to the underlying backend resources. If there is a conflict with backend-level tags, does not override them",
-          "title": "Tags",
-          "type": "object"
+          "title": "Tags"
         },
         "type": {
+          "const": "task",
           "default": "task",
-          "enum": [
-            "task"
-          ],
           "title": "Type",
           "type": "string"
         },
         "user": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The user inside the container, `user_name_or_id[:group_name_or_id]` (e.g., `ubuntu`, `1000:1000`). Defaults to the default user from the `image`",
-          "title": "User",
-          "type": "string"
+          "title": "User"
         },
         "utilization_policy": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/UtilizationPolicy"
+            },
             {
-              "$ref": "#/definitions/UtilizationPolicyRequest"
+              "type": "null"
             }
           ],
-          "description": "Run termination policy based on utilization",
-          "title": "Utilization Policy"
+          "default": null,
+          "description": "Run termination policy based on utilization"
         },
         "volumes": {
           "default": [],
@@ -3596,10 +4920,10 @@
           "items": {
             "anyOf": [
               {
-                "$ref": "#/definitions/VolumeMountPointRequest"
+                "$ref": "#/$defs/VolumeMountPoint"
               },
               {
-                "$ref": "#/definitions/InstanceMountPointRequest"
+                "$ref": "#/$defs/InstanceMountPoint"
               },
               {
                 "type": "string"
@@ -3610,15 +4934,23 @@
           "type": "array"
         },
         "working_dir": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The absolute path to the working directory inside the container. Defaults to the `image`'s default working directory",
-          "title": "Working Dir",
-          "type": "string"
+          "title": "Working Dir"
         }
       },
-      "title": "TaskConfigurationRequest",
+      "title": "TaskConfiguration",
       "type": "object"
     },
-    "UtilizationPolicyRequest": {
+    "UtilizationPolicy": {
       "additionalProperties": false,
       "properties": {
         "min_gpu_utilization": {
@@ -3645,11 +4977,10 @@
         "min_gpu_utilization",
         "time_window"
       ],
-      "title": "UtilizationPolicyRequest",
+      "title": "UtilizationPolicy",
       "type": "object"
     },
     "VastAIOfferOrder": {
-      "description": "An enumeration.",
       "enum": [
         "score",
         "price"
@@ -3661,31 +4992,49 @@
       "additionalProperties": false,
       "properties": {
         "min_reliability": {
+          "anyOf": [
+            {
+              "maximum": 1,
+              "minimum": 0,
+              "type": "number"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The minimum reliability threshold for offers, on a scale from `0` to `1`. Defaults to `0.9`",
-          "maximum": 1,
-          "minimum": 0,
-          "title": "Min Reliability",
-          "type": "number"
+          "title": "Min Reliability"
         },
         "min_score": {
+          "anyOf": [
+            {
+              "minimum": 0,
+              "type": "integer"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The minimum overall score required for offers to be considered. The scoring scale varies and may require experimentation. Starting with a value in the low hundreds is generally recommended",
-          "minimum": 0,
-          "title": "Min Score",
-          "type": "integer"
+          "title": "Min Score"
         },
         "offer_order": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/VastAIOfferOrder"
+            },
             {
-              "$ref": "#/definitions/VastAIOfferOrder"
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "Controls the order in which offers are considered for provisioning. Use `score` to prioritize the highest overall score first (the default order in the Vast.ai console), or `price` to prioritize the lowest-cost offers first. Lower-cost offers are often less reliable, so consider applying stricter filters when using `price`. Defaults to `score`"
         },
         "type": {
+          "const": "vastai",
           "default": "vastai",
-          "enum": [
-            "vastai"
-          ],
           "title": "Type",
           "type": "string"
         }
@@ -3693,34 +5042,33 @@
       "title": "VastAIProfileOptions",
       "type": "object"
     },
-    "VolumeConfigurationRequest": {
-      "additionalProperties": false,
+    "VolumeConfiguration": {
       "discriminator": {
         "mapping": {
-          "aws": "#/definitions/AWSVolumeConfigurationRequest",
-          "gcp": "#/definitions/GCPVolumeConfigurationRequest",
-          "kubernetes": "#/definitions/KubernetesVolumeConfigurationRequest",
-          "runpod": "#/definitions/RunpodVolumeConfigurationRequest"
+          "aws": "#/$defs/AWSVolumeConfiguration",
+          "gcp": "#/$defs/GCPVolumeConfiguration",
+          "kubernetes": "#/$defs/KubernetesVolumeConfiguration",
+          "runpod": "#/$defs/RunpodVolumeConfiguration"
         },
         "propertyName": "backend"
       },
       "oneOf": [
         {
-          "$ref": "#/definitions/AWSVolumeConfigurationRequest"
+          "$ref": "#/$defs/AWSVolumeConfiguration"
         },
         {
-          "$ref": "#/definitions/GCPVolumeConfigurationRequest"
+          "$ref": "#/$defs/GCPVolumeConfiguration"
         },
         {
-          "$ref": "#/definitions/RunpodVolumeConfigurationRequest"
+          "$ref": "#/$defs/RunpodVolumeConfiguration"
         },
         {
-          "$ref": "#/definitions/KubernetesVolumeConfigurationRequest"
+          "$ref": "#/$defs/KubernetesVolumeConfiguration"
         }
       ],
-      "title": "VolumeConfigurationRequest"
+      "title": "VolumeConfiguration"
     },
-    "VolumeMountPointRequest": {
+    "VolumeMountPoint": {
       "additionalProperties": false,
       "properties": {
         "name": {
@@ -3748,40 +5096,42 @@
         "name",
         "path"
       ],
-      "title": "VolumeMountPointRequest",
+      "title": "VolumeMountPoint",
       "type": "object"
     }
   },
+  "$schema": "https://json-schema.org/draft/2020-12/schema",
+  "additionalProperties": true,
   "discriminator": {
     "mapping": {
-      "dev-environment": "#/definitions/DevEnvironmentConfigurationRequest",
-      "fleet": "#/definitions/FleetConfigurationRequest",
-      "gateway": "#/definitions/GatewayConfigurationRequest",
-      "service": "#/definitions/ServiceConfigurationRequest",
-      "task": "#/definitions/TaskConfigurationRequest",
-      "volume": "#/definitions/VolumeConfigurationRequest"
+      "dev-environment": "#/$defs/DevEnvironmentConfiguration",
+      "fleet": "#/$defs/FleetConfiguration",
+      "gateway": "#/$defs/GatewayConfiguration",
+      "service": "#/$defs/ServiceConfiguration",
+      "task": "#/$defs/TaskConfiguration",
+      "volume": "#/$defs/VolumeConfiguration"
     },
     "propertyName": "type"
   },
   "oneOf": [
     {
-      "$ref": "#/definitions/DevEnvironmentConfigurationRequest"
+      "$ref": "#/$defs/DevEnvironmentConfiguration"
     },
     {
-      "$ref": "#/definitions/TaskConfigurationRequest"
+      "$ref": "#/$defs/TaskConfiguration"
     },
     {
-      "$ref": "#/definitions/ServiceConfigurationRequest"
+      "$ref": "#/$defs/ServiceConfiguration"
     },
     {
-      "$ref": "#/definitions/FleetConfigurationRequest"
+      "$ref": "#/$defs/FleetConfiguration"
     },
     {
-      "$ref": "#/definitions/GatewayConfigurationRequest"
+      "$ref": "#/$defs/GatewayConfiguration"
     },
     {
-      "$ref": "#/definitions/VolumeConfigurationRequest"
+      "$ref": "#/$defs/VolumeConfiguration"
     }
   ],
-  "title": "DstackConfigurationRequest"
+  "title": "DstackConfiguration"
 }
diff --git a/src/tests/_internal/pydantic_compat/fixtures/schema/profiles.json b/src/tests/_internal/pydantic_compat/fixtures/schema/profiles.json
index 9a71d9e472..94de590152 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/schema/profiles.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/schema/profiles.json
@@ -1,7 +1,5 @@
 {
-  "$schema": "http://json-schema.org/draft-07/schema#",
-  "additionalProperties": false,
-  "definitions": {
+  "$defs": {
     "BackendType": {
       "description": "Attributes:\n    AMDDEVCLOUD (BackendType): AMD Developer Cloud\n    AWS (BackendType): Amazon Web Services\n    AZURE (BackendType): Microsoft Azure\n    CLOUDRIFT (BackendType): CloudRift\n    CRUSOE (BackendType): Crusoe\n    CUDO (BackendType): Cudo\n    DATACRUNCH (BackendType): DataCrunch (for backward compatibility)\n    DIGITALOCEAN (BackendType): DigitalOcean\n    DSTACK (BackendType): dstack Sky\n    GCP (BackendType): Google Cloud Platform\n    HOTAISLE (BackendType): Hot Aisle\n    JARVISLABS (BackendType): JarvisLabs\n    KUBERNETES (BackendType): Kubernetes\n    LAMBDA (BackendType): Lambda Cloud\n    NEBIUS (BackendType): Nebius AI Cloud\n    OCI (BackendType): Oracle Cloud Infrastructure\n    RUNPOD (BackendType): Runpod Cloud\n    TENSORDOCK (BackendType): TensorDock Marketplace\n    VASTAI (BackendType): Vast.ai Marketplace\n    VERDA (BackendType): Verda Cloud\n    VULTR (BackendType): Vultr\n    SLURM (BackendType): Slurm",
       "enum": [
@@ -33,7 +31,6 @@
       "type": "string"
     },
     "CreationPolicy": {
-      "description": "An enumeration.",
       "enum": [
         "reuse",
         "reuse-or-create"
@@ -41,7 +38,7 @@
       "title": "CreationPolicy",
       "type": "string"
     },
-    "EntityReferenceRequest": {
+    "EntityReference": {
       "additionalProperties": false,
       "description": "Cross-project entity reference.",
       "properties": {
@@ -51,32 +48,38 @@
           "type": "string"
         },
         "project": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The project name. If unspecified, refers to the current project",
-          "title": "Project",
-          "type": "string"
+          "title": "Project"
         }
       },
       "required": [
         "name"
       ],
-      "title": "EntityReferenceRequest",
+      "title": "EntityReference",
       "type": "object"
     },
-    "FleetInstanceSelectorRequest": {
+    "FleetInstanceSelector": {
       "additionalProperties": false,
       "properties": {
         "fleet": {
           "anyOf": [
             {
-              "$ref": "#/definitions/EntityReferenceRequest"
+              "$ref": "#/$defs/EntityReference"
             },
             {
-              "minLength": 1,
               "type": "string"
             }
           ],
-          "description": "The fleet reference. For fleets owned by the current project, specify the fleet name. For a fleet from another project, specify `<project name>/<fleet name>` or an object with `project` and `name`.",
-          "title": "Fleet"
+          "description": "The fleet reference. For fleets owned by the current project, specify the fleet name. For a fleet from another project, specify `<project name>/<fleet name>` or an object with `project` and `name`."
         },
         "instance": {
           "description": "The fleet instance number",
@@ -89,10 +92,10 @@
         "fleet",
         "instance"
       ],
-      "title": "FleetInstanceSelectorRequest",
+      "title": "FleetInstanceSelector",
       "type": "object"
     },
-    "InstanceHostnameSelectorRequest": {
+    "InstanceHostnameSelector": {
       "additionalProperties": false,
       "properties": {
         "hostname": {
@@ -105,10 +108,10 @@
       "required": [
         "hostname"
       ],
-      "title": "InstanceHostnameSelectorRequest",
+      "title": "InstanceHostnameSelector",
       "type": "object"
     },
-    "InstanceNameSelectorRequest": {
+    "InstanceNameSelector": {
       "additionalProperties": false,
       "properties": {
         "name": {
@@ -121,41 +124,70 @@
       "required": [
         "name"
       ],
-      "title": "InstanceNameSelectorRequest",
+      "title": "InstanceNameSelector",
       "type": "object"
     },
-    "ProfileRequest": {
+    "Profile": {
       "additionalProperties": false,
       "properties": {
         "availability_zones": {
+          "anyOf": [
+            {
+              "items": {
+                "type": "string"
+              },
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The availability zones to consider for provisioning (e.g., `[eu-west-1a, us-west4-a]`)",
-          "items": {
-            "type": "string"
-          },
-          "title": "Availability Zones",
-          "type": "array"
+          "title": "Availability Zones"
         },
         "backend_options": {
+          "anyOf": [
+            {
+              "items": {
+                "$ref": "#/$defs/VastAIProfileOptions"
+              },
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "Backend-specific options, applied only to offers from that backend",
-          "items": {
-            "$ref": "#/definitions/VastAIProfileOptions"
-          },
-          "title": "Backend Options",
-          "type": "array"
+          "title": "Backend Options"
         },
         "backends": {
+          "anyOf": [
+            {
+              "items": {
+                "$ref": "#/$defs/BackendType"
+              },
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The backends to consider for provisioning (e.g., `[aws, gcp]`)",
-          "items": {
-            "$ref": "#/definitions/BackendType"
-          },
-          "type": "array"
+          "title": "Backends"
         },
         "creation_policy": {
-          "allOf": [
+          "anyOf": [
             {
-              "$ref": "#/definitions/CreationPolicy"
+              "$ref": "#/$defs/CreationPolicy"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The policy for using instances from fleets: `reuse`, `reuse-or-create`. Defaults to `reuse-or-create`"
         },
         "default": {
@@ -165,89 +197,129 @@
           "type": "boolean"
         },
         "fleets": {
-          "description": "The fleets considered for reuse. For fleets owned by the current project, specify fleet names. For imported fleets, specify `<project name>/<fleet name>`",
-          "items": {
-            "anyOf": [
-              {
-                "$ref": "#/definitions/EntityReferenceRequest"
+          "anyOf": [
+            {
+              "items": {
+                "anyOf": [
+                  {
+                    "$ref": "#/$defs/EntityReference"
+                  },
+                  {
+                    "type": "string"
+                  }
+                ]
               },
-              {
-                "type": "string"
-              }
-            ]
-          },
-          "title": "Fleets",
-          "type": "array"
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
+          "description": "The fleets considered for reuse. For fleets owned by the current project, specify fleet names. For imported fleets, specify `<project name>/<fleet name>`",
+          "title": "Fleets"
         },
         "idle_duration": {
           "anyOf": [
+            {
+              "const": "off",
+              "type": "string"
+            },
             {
               "type": "integer"
             },
             {
               "type": "string"
+            },
+            {
+              "type": "boolean"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "Time to wait before terminating idle instances. When the run reuses an existing fleet instance, the fleet's `idle_duration` applies. When the run provisions a new instance, the shorter of the fleet's and run's values is used. Defaults to `5m` for runs and `3d` for fleets. Use `off` for unlimited duration. Only applied for VM-based backends",
           "title": "Idle Duration"
         },
         "instance_types": {
+          "anyOf": [
+            {
+              "items": {
+                "type": "string"
+              },
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The cloud-specific instance types to consider for provisioning (e.g., `[g6e.24xlarge, n1-standard-4]`)",
-          "items": {
-            "type": "string"
-          },
-          "title": "Instance Types",
-          "type": "array"
+          "title": "Instance Types"
         },
         "instances": {
-          "description": "The specific fleet instances to consider for reuse. Each value can be an instance name string, or an object with `name`, `hostname`, or `fleet` and `instance`. When set, the run is only placed on matching existing instances.",
-          "items": {
-            "anyOf": [
-              {
-                "$ref": "#/definitions/InstanceNameSelectorRequest"
-              },
-              {
-                "$ref": "#/definitions/InstanceHostnameSelectorRequest"
-              },
-              {
-                "$ref": "#/definitions/FleetInstanceSelectorRequest"
+          "anyOf": [
+            {
+              "items": {
+                "anyOf": [
+                  {
+                    "$ref": "#/$defs/InstanceNameSelector"
+                  },
+                  {
+                    "$ref": "#/$defs/InstanceHostnameSelector"
+                  },
+                  {
+                    "$ref": "#/$defs/FleetInstanceSelector"
+                  },
+                  {
+                    "type": "string"
+                  }
+                ]
               },
-              {
-                "minLength": 1,
-                "type": "string"
-              }
-            ]
-          },
-          "minItems": 1,
-          "title": "Instances",
-          "type": "array"
+              "minItems": 1,
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
+          "description": "The specific fleet instances to consider for reuse. Each value can be an instance name string, or an object with `name`, `hostname`, or `fleet` and `instance`. When set, the run is only placed on matching existing instances.",
+          "title": "Instances"
         },
         "max_duration": {
           "anyOf": [
             {
-              "enum": [
-                "off"
-              ],
-              "type": "string"
+              "type": "integer"
             },
             {
-              "type": "integer"
+              "type": "string"
             },
             {
               "type": "boolean"
             },
             {
-              "type": "string"
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The maximum duration of a run (e.g., `2h`, `1d`, etc) in a running state, excluding provisioning and pulling. After it elapses, the run is automatically stopped. Use `off` for unlimited duration. Defaults to `off`",
           "title": "Max Duration"
         },
         "max_price": {
+          "anyOf": [
+            {
+              "exclusiveMinimum": 0.0,
+              "type": "number"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The maximum instance price per hour, in dollars",
-          "exclusiveMinimum": 0.0,
-          "title": "Max Price",
-          "type": "number"
+          "title": "Max Price"
         },
         "name": {
           "default": "",
@@ -256,106 +328,150 @@
           "type": "string"
         },
         "regions": {
+          "anyOf": [
+            {
+              "items": {
+                "type": "string"
+              },
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The regions to consider for provisioning (e.g., `[eu-west-1, us-west4, westeurope]`)",
-          "items": {
-            "type": "string"
-          },
-          "title": "Regions",
-          "type": "array"
+          "title": "Regions"
         },
         "reservation": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The existing reservation to use for instance provisioning. Supports AWS Capacity Reservations, AWS Capacity Blocks, and GCP reservations",
-          "title": "Reservation",
-          "type": "string"
+          "title": "Reservation"
         },
         "retry": {
           "anyOf": [
             {
-              "$ref": "#/definitions/ProfileRetryRequest"
+              "$ref": "#/$defs/ProfileRetry"
             },
             {
               "type": "boolean"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The policy for resubmitting the run. Defaults to `false`",
           "title": "Retry"
         },
         "schedule": {
-          "allOf": [
+          "anyOf": [
             {
-              "$ref": "#/definitions/ScheduleRequest"
+              "$ref": "#/$defs/Schedule"
+            },
+            {
+              "type": "null"
             }
           ],
-          "description": "The schedule for starting the run at specified time",
-          "title": "Schedule"
+          "default": null,
+          "description": "The schedule for starting the run at specified time"
         },
         "spot_policy": {
-          "allOf": [
+          "anyOf": [
             {
-              "$ref": "#/definitions/SpotPolicy"
+              "$ref": "#/$defs/SpotPolicy"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The policy for provisioning spot or on-demand instances: `spot`, `on-demand`, `auto`. Defaults to `on-demand`"
         },
         "startup_order": {
-          "allOf": [
+          "anyOf": [
             {
-              "$ref": "#/definitions/StartupOrder"
+              "$ref": "#/$defs/StartupOrder"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The order in which master and workers jobs are started: `any`, `master-first`, `workers-first`. Defaults to `any`"
         },
         "stop_criteria": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/StopCriteria"
+            },
             {
-              "$ref": "#/definitions/StopCriteria"
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The criteria determining when a multi-node run should be considered finished: `all-done`, `master-done`. Defaults to `all-done`"
         },
         "stop_duration": {
           "anyOf": [
             {
-              "enum": [
-                "off"
-              ],
-              "type": "string"
+              "type": "integer"
             },
             {
-              "type": "integer"
+              "type": "string"
             },
             {
               "type": "boolean"
             },
             {
-              "type": "string"
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The maximum duration of a run graceful stopping. After it elapses, the run is automatically forced stopped. This includes force detaching volumes used by the run. Use `off` for unlimited duration. Defaults to `5m`",
           "title": "Stop Duration"
         },
         "tags": {
-          "additionalProperties": {
-            "type": "string"
-          },
+          "anyOf": [
+            {
+              "additionalProperties": {
+                "type": "string"
+              },
+              "type": "object"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The custom tags to associate with the resource. The tags are also propagated to the underlying backend resources. If there is a conflict with backend-level tags, does not override them",
-          "title": "Tags",
-          "type": "object"
+          "title": "Tags"
         },
         "utilization_policy": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/UtilizationPolicy"
+            },
             {
-              "$ref": "#/definitions/UtilizationPolicyRequest"
+              "type": "null"
             }
           ],
-          "description": "Run termination policy based on utilization",
-          "title": "Utilization Policy"
+          "default": null,
+          "description": "Run termination policy based on utilization"
         }
       },
-      "title": "ProfileRequest",
+      "title": "Profile",
       "type": "object"
     },
-    "ProfileRetryRequest": {
+    "ProfileRetry": {
       "additionalProperties": false,
       "properties": {
         "duration": {
@@ -365,24 +481,36 @@
             },
             {
               "type": "string"
+            },
+            {
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "The maximum period of retrying the run, e.g., `4h` or `1d`. The period is calculated as a run age for `no-capacity` event and as a time passed since the last `interruption` and `error` for `interruption` and `error` events.",
           "title": "Duration"
         },
         "on_events": {
+          "anyOf": [
+            {
+              "items": {
+                "$ref": "#/$defs/RetryEvent"
+              },
+              "type": "array"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The list of events that should be handled with retry. Supported events are `no-capacity`, `interruption`, `error`. Omit to retry on all events",
-          "items": {
-            "$ref": "#/definitions/RetryEvent"
-          },
-          "type": "array"
+          "title": "On Events"
         }
       },
-      "title": "ProfileRetryRequest",
+      "title": "ProfileRetry",
       "type": "object"
     },
     "RetryEvent": {
-      "description": "An enumeration.",
       "enum": [
         "no-capacity",
         "interruption",
@@ -391,7 +519,7 @@
       "title": "RetryEvent",
       "type": "string"
     },
-    "ScheduleRequest": {
+    "Schedule": {
       "additionalProperties": false,
       "properties": {
         "cron": {
@@ -413,11 +541,10 @@
       "required": [
         "cron"
       ],
-      "title": "ScheduleRequest",
+      "title": "Schedule",
       "type": "object"
     },
     "SpotPolicy": {
-      "description": "An enumeration.",
       "enum": [
         "spot",
         "on-demand",
@@ -427,7 +554,6 @@
       "type": "string"
     },
     "StartupOrder": {
-      "description": "An enumeration.",
       "enum": [
         "any",
         "master-first",
@@ -437,7 +563,6 @@
       "type": "string"
     },
     "StopCriteria": {
-      "description": "An enumeration.",
       "enum": [
         "all-done",
         "master-done"
@@ -445,7 +570,7 @@
       "title": "StopCriteria",
       "type": "string"
     },
-    "UtilizationPolicyRequest": {
+    "UtilizationPolicy": {
       "additionalProperties": false,
       "properties": {
         "min_gpu_utilization": {
@@ -472,11 +597,10 @@
         "min_gpu_utilization",
         "time_window"
       ],
-      "title": "UtilizationPolicyRequest",
+      "title": "UtilizationPolicy",
       "type": "object"
     },
     "VastAIOfferOrder": {
-      "description": "An enumeration.",
       "enum": [
         "score",
         "price"
@@ -488,31 +612,49 @@
       "additionalProperties": false,
       "properties": {
         "min_reliability": {
+          "anyOf": [
+            {
+              "maximum": 1,
+              "minimum": 0,
+              "type": "number"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The minimum reliability threshold for offers, on a scale from `0` to `1`. Defaults to `0.9`",
-          "maximum": 1,
-          "minimum": 0,
-          "title": "Min Reliability",
-          "type": "number"
+          "title": "Min Reliability"
         },
         "min_score": {
+          "anyOf": [
+            {
+              "minimum": 0,
+              "type": "integer"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
           "description": "The minimum overall score required for offers to be considered. The scoring scale varies and may require experimentation. Starting with a value in the low hundreds is generally recommended",
-          "minimum": 0,
-          "title": "Min Score",
-          "type": "integer"
+          "title": "Min Score"
         },
         "offer_order": {
-          "allOf": [
+          "anyOf": [
+            {
+              "$ref": "#/$defs/VastAIOfferOrder"
+            },
             {
-              "$ref": "#/definitions/VastAIOfferOrder"
+              "type": "null"
             }
           ],
+          "default": null,
           "description": "Controls the order in which offers are considered for provisioning. Use `score` to prioritize the highest overall score first (the default order in the Vast.ai console), or `price` to prioritize the lowest-cost offers first. Lower-cost offers are often less reliable, so consider applying stricter filters when using `price`. Defaults to `score`"
         },
         "type": {
+          "const": "vastai",
           "default": "vastai",
-          "enum": [
-            "vastai"
-          ],
           "title": "Type",
           "type": "string"
         }
@@ -521,10 +663,12 @@
       "type": "object"
     }
   },
+  "$schema": "https://json-schema.org/draft/2020-12/schema",
+  "additionalProperties": false,
   "properties": {
     "profiles": {
       "items": {
-        "$ref": "#/definitions/ProfileRequest"
+        "$ref": "#/$defs/Profile"
       },
       "title": "Profiles",
       "type": "array"
@@ -533,6 +677,6 @@
   "required": [
     "profiles"
   ],
-  "title": "ProfilesConfigRequest",
+  "title": "ProfilesConfig",
   "type": "object"
 }
diff --git a/src/tests/_internal/pydantic_compat/fixtures/serialization/api_response/fleet.json b/src/tests/_internal/pydantic_compat/fixtures/serialization/api_response/fleet.json
index 2b0b2d6f78..ead4a09501 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/serialization/api_response/fleet.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/serialization/api_response/fleet.json
@@ -1,15 +1,16 @@
 {
-  "created_at": "2024-01-02T03:04:05+00:00",
+  "created_at": "2024-01-02T03:04:05Z",
   "id": "11111111-1111-4111-8111-111111111111",
   "instances": [
     {
       "availability_zone": null,
       "backend": "aws",
       "busy_blocks": 0,
-      "created": "2024-01-02T03:04:05+00:00",
+      "created": "2024-01-02T03:04:05Z",
       "finished_at": null,
       "fleet_id": null,
       "fleet_name": null,
+      "gpu_driver": null,
       "health_status": "healthy",
       "hostname": null,
       "id": "11111111-1111-4111-8111-111111111111",
diff --git a/src/tests/_internal/pydantic_compat/fixtures/serialization/api_response/gateway.json b/src/tests/_internal/pydantic_compat/fixtures/serialization/api_response/gateway.json
index 0a6bee732b..3cacb9291b 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/serialization/api_response/gateway.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/serialization/api_response/gateway.json
@@ -16,7 +16,7 @@
     "tags": null,
     "type": "gateway"
   },
-  "created_at": "2024-01-02T03:04:05+00:00",
+  "created_at": "2024-01-02T03:04:05Z",
   "default": true,
   "hostname": "gateway.example.com",
   "id": "11111111-1111-4111-8111-111111111111",
diff --git a/src/tests/_internal/pydantic_compat/fixtures/serialization/api_response/project.json b/src/tests/_internal/pydantic_compat/fixtures/serialization/api_response/project.json
index b4a6f0a4c3..3e0c4e4b73 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/serialization/api_response/project.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/serialization/api_response/project.json
@@ -11,7 +11,7 @@
       "project_role": "admin",
       "user": {
         "active": true,
-        "created_at": "2024-01-02T03:04:05+00:00",
+        "created_at": "2024-01-02T03:04:05Z",
         "email": null,
         "global_role": "user",
         "id": "11111111-1111-4111-8111-111111111111",
@@ -25,7 +25,7 @@
   ],
   "owner": {
     "active": true,
-    "created_at": "2024-01-02T03:04:05+00:00",
+    "created_at": "2024-01-02T03:04:05Z",
     "email": null,
     "global_role": "user",
     "id": "11111111-1111-4111-8111-111111111111",
diff --git a/src/tests/_internal/pydantic_compat/fixtures/serialization/api_response/user_with_creds.json b/src/tests/_internal/pydantic_compat/fixtures/serialization/api_response/user_with_creds.json
index 97c4e2f379..f2d67619e0 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/serialization/api_response/user_with_creds.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/serialization/api_response/user_with_creds.json
@@ -1,6 +1,6 @@
 {
   "active": true,
-  "created_at": "2024-01-02T03:04:05+00:00",
+  "created_at": "2024-01-02T03:04:05Z",
   "creds": {
     "token": "test-token"
   },
diff --git a/src/tests/_internal/pydantic_compat/fixtures/serialization/api_response/volume.json b/src/tests/_internal/pydantic_compat/fixtures/serialization/api_response/volume.json
index 80ec497bad..c4b59fd8f8 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/serialization/api_response/volume.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/serialization/api_response/volume.json
@@ -12,13 +12,13 @@
     "type": "volume",
     "volume_id": null
   },
-  "cost": 0,
-  "created_at": "2024-01-02T03:04:05+00:00",
+  "cost": 0.0,
+  "created_at": "2024-01-02T03:04:05Z",
   "deleted": false,
   "deleted_at": null,
   "external": false,
   "id": "11111111-1111-4111-8111-111111111111",
-  "last_processed_at": "2024-01-02T03:04:05+00:00",
+  "last_processed_at": "2024-01-02T03:04:05Z",
   "name": "test-volume",
   "project_name": "test-project",
   "provisioning_data": null,
diff --git a/src/tests/_internal/pydantic_compat/fixtures/serialization/db/job_provisioning_data.json b/src/tests/_internal/pydantic_compat/fixtures/serialization/db/job_provisioning_data.json
index 6f8c681be2..81f60576d1 100644
--- a/src/tests/_internal/pydantic_compat/fixtures/serialization/db/job_provisioning_data.json
+++ b/src/tests/_internal/pydantic_compat/fixtures/serialization/db/job_provisioning_data.json
@@ -4,6 +4,7 @@
   "backend_data": null,
   "base_backend": null,
   "dockerized": true,
+  "gpu_driver": null,
   "hostname": "127.0.0.4",
   "instance_id": "instance_id",
   "instance_network": null,
diff --git a/src/tests/_internal/pydantic_compat/test_custom_types.py b/src/tests/_internal/pydantic_compat/test_custom_types.py
index cdcade6556..97f6a3ec79 100644
--- a/src/tests/_internal/pydantic_compat/test_custom_types.py
+++ b/src/tests/_internal/pydantic_compat/test_custom_types.py
@@ -11,9 +11,9 @@
 from typing import Any, Callable
 
 import pytest
-from pydantic import ValidationError, parse_obj_as
+from pydantic import TypeAdapter, ValidationError
 
-from dstack._internal.core.models.common import Duration
+from dstack._internal.core.models.duration import Duration
 from dstack._internal.core.models.gateways import GatewaySpec
 from dstack._internal.core.models.resources import (
     ComputeCapability,
@@ -92,19 +92,19 @@ def test_specialization_preserves_spec_type_and_production_request_body(
     ):
         """
         Pydantic v1 represents these specializations as typing aliases and adds
-        `__orig_class__` after construction. Production sends `.dict()` to requests, relying on
+        `__orig_class__` after construction. Production sends `.model_dump()` to requests, relying on
         the request model's override to remove that non-JSON value.
         """
         spec = spec_factory()
         request = request_model(user="alice", project="main", spec=spec)
 
         assert class_name(request.spec) == expected_spec_type
-        body = request.dict()
+        body = request.model_dump()
         assert "__orig_class__" not in body
 
         # Match the exact production handoff: requests receives a plain, stdlib-JSON-safe dict.
         encoded_body = json.dumps(body)
-        assert canonicalize(json.dumps(body["spec"])) == canonicalize(spec.json())
+        assert canonicalize(json.dumps(body["spec"])) == canonicalize(spec.model_dump_json())
 
         # The generic must also choose the same type when its value arrives as an untyped payload.
         reparsed_request = request_model(**json.loads(encoded_body))
@@ -160,24 +160,24 @@ def test_specialization_preserves_bound_types_and_json(
         expected: dict[str, Any],
         bound_type: type,
     ):
-        value = parse_obj_as(range_type, raw)
+        value = TypeAdapter(range_type).validate_python(raw)
 
         assert type(value) is range_type
-        assert value.dict() == expected
+        assert value.model_dump() == expected
         for bound in (value.min, value.max):
             if bound is not None:
                 assert type(bound) is bound_type
-        assert canonicalize(value.json()) == canonicalize(json.dumps(expected))
+        assert canonicalize(value.model_dump_json()) == canonicalize(json.dumps(expected))
 
     @pytest.mark.parametrize("raw", ["..", "8..2", "1...3"])
     def test_invalid_int_ranges_stay_rejected(self, raw: str):
         with pytest.raises(ValidationError):
-            parse_obj_as(Range[int], raw)
+            Range[int].model_validate(raw)
 
     @pytest.mark.parametrize("raw", ["...", "2TB..1TB"])
     def test_invalid_memory_ranges_stay_rejected(self, raw: str):
         with pytest.raises(ValidationError):
-            parse_obj_as(Range[Memory], raw)
+            Range[Memory].model_validate(raw)
 
 
 _SCALAR_CASES = [
@@ -239,7 +239,7 @@ def test_parsed_value_type_and_json_stay_stable(
         expected_type: type,
         expected_json: Any,
     ):
-        value = parse_obj_as(scalar_type, raw)
+        value = TypeAdapter(scalar_type).validate_python(raw)
 
         assert value == expected
         assert type(value) is expected_type
@@ -258,7 +258,7 @@ def test_parsed_value_type_and_json_stay_stable(
     )
     def test_invalid_values_stay_rejected(self, scalar_type: Any, raw: Any):
         with pytest.raises(ValidationError):
-            parse_obj_as(scalar_type, raw)
+            TypeAdapter(scalar_type).validate_python(raw)
 
 
 _CUSTOM_MODEL_CASES = [
@@ -391,9 +391,9 @@ def test_custom_parser_preserves_values_types_and_json(
         expected_json: dict[str, Any],
         expected_types: dict[str, str],
     ):
-        value = parse_obj_as(model_type, raw)
+        value = TypeAdapter(model_type).validate_python(raw)
 
-        assert canonicalize(value.json()) == canonicalize(json.dumps(expected_json))
+        assert canonicalize(value.model_dump_json()) == canonicalize(json.dumps(expected_json))
         assert type_map(value) == expected_types
 
     @pytest.mark.parametrize(
@@ -408,4 +408,4 @@ def test_custom_parser_preserves_values_types_and_json(
     )
     def test_invalid_custom_model_values_stay_rejected(self, model_type: Any, raw: Any):
         with pytest.raises(ValidationError):
-            parse_obj_as(model_type, raw)
+            TypeAdapter(model_type).validate_python(raw)
diff --git a/src/tests/_internal/pydantic_compat/test_field_parsing.py b/src/tests/_internal/pydantic_compat/test_field_parsing.py
index dabf94e81a..6fe9aaa46f 100644
--- a/src/tests/_internal/pydantic_compat/test_field_parsing.py
+++ b/src/tests/_internal/pydantic_compat/test_field_parsing.py
@@ -7,7 +7,6 @@
 import yaml
 
 from dstack._internal.core.errors import ConfigurationError
-from dstack._internal.core.models.common import Duration
 from dstack._internal.core.models.configurations import (
     PythonVersion,
     parse_apply_configuration,
@@ -39,7 +38,7 @@ def test_yaml_310_float_is_recovered_as_python_310(self):
         config = parse_apply_configuration(data)
 
         assert config.python is PythonVersion.PY310
-        assert json.loads(config.json())["python"] == "3.10"
+        assert json.loads(config.model_dump_json())["python"] == "3.10"
 
     def test_yaml_311_float_stays_python_311(self):
         data = yaml.safe_load(
@@ -63,7 +62,7 @@ def test_list_is_normalized_to_mapping_and_missing_value_becomes_sentinel(self):
         assert config.env["EMPTY"] == ""
         assert config.env["B"] == EnvSentinel(key="B")
         assert class_name(config.env["B"]) == "EnvSentinel"
-        assert json.loads(config.json())["env"] == {
+        assert json.loads(config.model_dump_json())["env"] == {
             "A": "1",
             "B": {"key": "B"},
             "EMPTY": "",
@@ -86,7 +85,7 @@ def test_task_port_variants_are_normalized_to_port_mappings(self):
         config = parse_apply_configuration(_task(ports=[8080, "8081:81", "*:82"]))
 
         assert all(class_name(port) == "PortMapping" for port in config.ports)
-        assert [port.dict() for port in config.ports] == [
+        assert [port.model_dump() for port in config.ports] == [
             {"local_port": 8080, "container_port": 8080},
             {"local_port": 8081, "container_port": 81},
             {"local_port": None, "container_port": 82},
@@ -111,7 +110,7 @@ def test_service_port_variants_are_normalized(self, raw: Any, expected: dict[str
         config = parse_apply_configuration(_service(port=raw))
 
         assert class_name(config.port) == "PortMapping"
-        assert config.port.dict() == expected
+        assert config.port.model_dump() == expected
 
 
 class TestMountPointFieldParsing:
@@ -121,9 +120,9 @@ def test_string_arms_select_volume_and_instance_mount_models(self):
         )
 
         assert class_name(config.volumes[0]) == "VolumeMountPoint"
-        assert config.volumes[0].dict() == {"name": "my-volume", "path": "/mnt/data"}
+        assert config.volumes[0].model_dump() == {"name": "my-volume", "path": "/mnt/data"}
         assert class_name(config.volumes[1]) == "InstanceMountPoint"
-        assert config.volumes[1].dict() == {
+        assert config.volumes[1].model_dump() == {
             "instance_path": "/host/cache",
             "path": "/cache",
             "optional": False,
@@ -142,7 +141,7 @@ def test_unix_and_windows_sources_keep_colons_in_the_source_path(self):
         )
 
         assert all(class_name(mapping) == "FilePathMapping" for mapping in config.files)
-        assert [mapping.dict() for mapping in config.files] == [
+        assert [mapping.model_dump() for mapping in config.files] == [
             {"local_path": "data", "path": "/workspace/data"},
             {"local_path": r"C:\data", "path": "/workspace/windows"},
         ]
@@ -230,7 +229,7 @@ def test_user_field_runs_unix_user_validation_without_changing_wire_value(
         config = parse_apply_configuration(_task(user=raw))
 
         assert config.user == raw
-        assert UnixUser.parse(config.user).dict() == parsed
+        assert UnixUser.parse(config.user).model_dump() == parsed
 
     @pytest.mark.parametrize(
         "raw",
@@ -253,7 +252,10 @@ def test_invalid_user_stays_rejected(self, raw: str):
     pytest.param("idle_duration", "off", -1, int, id="idle-off"),
     pytest.param("idle_duration", False, -1, int, id="idle-false"),
     pytest.param("idle_duration", -1, -1, int, id="idle-legacy-minus-one"),
-    pytest.param("max_duration", "2h", 7200, Duration, id="max-duration"),
+    # `int`, not `Duration`: the field is declared `Union[Literal["off"], int]`, so the `Duration`
+    # int-subclass the before-validator returns does not survive validation. Value and wire format
+    # are unaffected, and nothing does `isinstance(..., Duration)`.
+    pytest.param("max_duration", "2h", 7200, int, id="max-duration"),
 ]
 
 
@@ -291,7 +293,7 @@ def test_string_shorthand_becomes_openai_model(self):
         config = parse_apply_configuration(_service(model="llama"))
 
         assert class_name(config.model) == "OpenAIChatModel"
-        assert config.model.dict() == {
+        assert config.model.model_dump() == {
             "type": "chat",
             "name": "llama",
             "format": "openai",
@@ -320,7 +322,36 @@ def test_project_qualified_string_becomes_entity_reference(self):
         config = parse_apply_configuration(_service(gateway="other-project/shared-gateway"))
 
         assert class_name(config.gateway) == "EntityReference"
-        assert config.gateway.dict() == {
+        assert config.gateway.model_dump() == {
             "project": "other-project",
             "name": "shared-gateway",
         }
+
+    def test_bare_string_becomes_entity_reference(self):
+        config = parse_apply_configuration(_service(gateway="shared-gateway"))
+
+        assert class_name(config.gateway) == "EntityReference"
+        assert config.gateway.model_dump() == {"project": None, "name": "shared-gateway"}
+
+    @pytest.mark.parametrize(
+        ("value", "expected"),
+        [
+            (True, True),
+            (False, False),
+            # Quoted in YAML, so the field receives a `str`. The `bool` arm must still claim it:
+            # the union also has a `str` arm, which pydantic v2's "smart" mode prefers for an exact
+            # type match, turning `gateway: "false"` into a gateway *named* `false`.
+            ("true", True),
+            ("false", False),
+            ("yes", True),
+            ("no", False),
+            ("on", True),
+            ("off", False),
+            ("1", True),
+            ("0", False),
+        ],
+    )
+    def test_boolean_and_quoted_boolean_stay_boolean(self, value: Any, expected: bool):
+        config = parse_apply_configuration(_service(gateway=value))
+
+        assert config.gateway is expected
diff --git a/src/tests/_internal/pydantic_compat/test_parsing.py b/src/tests/_internal/pydantic_compat/test_parsing.py
index f3c7c59093..087e3e1cf1 100644
--- a/src/tests/_internal/pydantic_compat/test_parsing.py
+++ b/src/tests/_internal/pydantic_compat/test_parsing.py
@@ -78,7 +78,7 @@ def _derived_registry(surface: str) -> dict[str, Any]:
 
 # The `config` column is written as `XStoredConfig` but read back as `XConfig`, so this is the one
 # surface whose parse target is a different class from the one that produced the bytes. The read is
-# a splice of two columns — `XConfig(**json.loads(config), creds=XCreds.parse_raw(auth))` — and the
+# a splice of two columns — `XConfig(**json.loads(config), creds=XCreds.model_validate_json(auth))` — and the
 # inputs mirror that by carrying `creds` inline, which additionally makes the creds union resolve
 # here rather than arrive pre-resolved.
 BACKEND_CONFIGS = backend_factories.BACKEND_CONFIG_MODELS
@@ -93,7 +93,7 @@ def _derived_registry(surface: str) -> dict[str, Any]:
     "dev_environment": parse_apply_configuration,
     "fleet": parse_apply_configuration,
     "gateway": parse_apply_configuration,
-    "profiles": ProfilesConfig.parse_obj,
+    "profiles": ProfilesConfig.model_validate,
     "service": parse_apply_configuration,
     "task": parse_apply_configuration,
     "volume": parse_apply_configuration,
@@ -167,7 +167,7 @@ def test_resolves_the_intended_union_arm(self, name):
         expected = class_name(backend_factories.CREDS_ARMS[name])
         # Compared by name rather than `isinstance`: the permissive read yields duality's
         # `...Response` variant on v1 and the plain class on v2, and `class_name` erases that.
-        assert class_name(creds.__root__) == expected
+        assert class_name(creds.root) == expected
 
 
 class TestBackendDataParsing:
@@ -179,7 +179,7 @@ def test_parses_to_expected_values_and_types(self, name, regen):
 
 class TestNebiusOfferBackendDataSetField:
     """
-    Excluded from `BACKEND_DATA` because `.json()` raises on its `set` field, so there is nothing
+    Excluded from `BACKEND_DATA` because `.model_dump_json()` raises on its `set` field, so there is nothing
     for `_assert_parses` to compare. The read path is real regardless, so assert on the value.
     """
 
@@ -234,7 +234,9 @@ def test_unknown_field_is_dropped(self, surface, name):
         payload = _load_input(surface, name)
         baseline = parse_ignore_extra(model, payload)
         perturbed = parse_ignore_extra(model, {**payload, "unknown_from_a_newer_writer": {"x": 1}})
-        assert canonicalize(perturbed.json()) == canonicalize(baseline.json())
+        assert canonicalize(perturbed.model_dump_json()) == canonicalize(
+            baseline.model_dump_json()
+        )
 
 
 class TestUnknownFieldRejection:
@@ -311,7 +313,7 @@ def test_parses_to_expected_values_and_types(self, name, regen):
 
 def _assert_parses(surface: str, name: str, model, regen: bool) -> None:
     kind = f"parsing/{surface}"
-    assert_matches_fixture(kind, f"{name}.values", model.json(), regen=regen)
+    assert_matches_fixture(kind, f"{name}.values", model.model_dump_json(), regen=regen)
     assert_matches_fixture(kind, f"{name}.types", json.dumps(type_map(model)), regen=regen)
 
 
diff --git a/src/tests/_internal/pydantic_compat/test_rejection.py b/src/tests/_internal/pydantic_compat/test_rejection.py
index 2099615dd5..0eab7f5e1a 100644
--- a/src/tests/_internal/pydantic_compat/test_rejection.py
+++ b/src/tests/_internal/pydantic_compat/test_rejection.py
@@ -39,7 +39,7 @@ def _task(extra_yaml: str) -> Any:
         ConfigurationError,
     ),
     "request_unknown_key": (
-        CreateVolumeRequest.parse_obj,
+        CreateVolumeRequest.model_validate,
         {
             "configuration": {"type": "volume", "name": "v", "backend": "aws", "region": "r"},
             "unexpected": 1,
diff --git a/src/tests/_internal/pydantic_compat/test_schema.py b/src/tests/_internal/pydantic_compat/test_schema.py
index 4853558377..26aacf525e 100644
--- a/src/tests/_internal/pydantic_compat/test_schema.py
+++ b/src/tests/_internal/pydantic_compat/test_schema.py
@@ -38,7 +38,7 @@ class TestPublishedSchemas:
     @pytest.mark.parametrize("name", sorted(PUBLISHED_SCHEMAS))
     def test_matches_fixture(self, name, regen):
         # The exact call the CI job makes.
-        schema_json = PUBLISHED_SCHEMAS[name].schema_json()
+        schema_json = json.dumps(PUBLISHED_SCHEMAS[name].model_json_schema())
         assert_matches_fixture("schema", name, schema_json, regen=regen)
 
     @pytest.mark.parametrize("name", sorted(PUBLISHED_SCHEMAS))
@@ -51,7 +51,7 @@ def test_every_ref_resolves(self, name):
         This is not hypothetical: `add_extra_schema_types` in `utils/json_schema.py` rewrites
         properties in place and has already produced a `KeyError: '$ref'` in this job once.
         """
-        schema = json.loads(PUBLISHED_SCHEMAS[name].schema_json())
+        schema = PUBLISHED_SCHEMAS[name].model_json_schema()
         definitions = _definitions(schema)
         assert definitions, "expected the schema to define types to reference"
         unresolved = sorted(
diff --git a/src/tests/_internal/pydantic_compat/test_serialization.py b/src/tests/_internal/pydantic_compat/test_serialization.py
index 2669ad1c51..efc51f4434 100644
--- a/src/tests/_internal/pydantic_compat/test_serialization.py
+++ b/src/tests/_internal/pydantic_compat/test_serialization.py
@@ -5,8 +5,8 @@
 each registry below — and compared against a fixture generated under pydantic v1. On v1 these are
 regression tests; on the v2 branch they are the compat assertion.
 
-Nothing here may reference the duality API (`__request__` / `__response__`): these tests have to
-run unchanged on both versions, and duality is gone in v2.
+Nothing here may reference the duality request/response variants directly: these tests have to run
+unchanged on both versions, and duality is gone in v2. See `compat.py`.
 
 Disposable: this package is deleted once the v2 release is verified in prod, except for a curated
 subset of the `db/` fixtures, which outlive it because stored rows do.
@@ -28,7 +28,7 @@
 from dstack._internal.core.models.repos.remote import RemoteRunRepoData
 from dstack._internal.core.models.resources import CPUSpec, Range, ResourcesSpec
 from dstack._internal.core.models.volumes import RunpodVolumeConfiguration, VolumeSpec
-from dstack._internal.server.utils.routers import CustomORJSONResponse
+from dstack._internal.server.utils.routers import CustomJSONResponse
 from tests._internal.pydantic_compat import backend_factories, factories
 from tests._internal.pydantic_compat.compare import assert_matches_fixture, canonicalize
 
@@ -75,7 +75,7 @@
     "save_repo_creds_request": factories.save_repo_creds_request,
 }
 
-# Returned from a router via `CustomORJSONResponse` — orjson with a `default=` hook rather than
+# Returned from a router via `CustomJSONResponse` — orjson with a `default=` hook rather than
 # `.json()`, so this path can drift away from the one above.
 API_RESPONSES: dict[str, Callable[[], CoreModel]] = {
     "fleet": factories.fleet,
@@ -89,7 +89,8 @@
     "volume": factories.volume,
 }
 
-# Sent by the server to the runner (shim) as a request body, via `.json()`.
+# Sent by the server to the runner (shim) as a request body. `SubmitBody` restricts what it
+# sends via `json_for_runner()`; the rest go out as plain `.json()`.
 RUNNER_REQUESTS: dict[str, Callable[[], CoreModel]] = {
     "component_install_request": factories.component_install_request,
     "legacy_submit_body": factories.legacy_submit_body,
@@ -127,16 +128,21 @@
 # unknown-field tolerance: it has to use the identical path, or it compares orjson output against
 # `.json()` output and fails for reasons that have nothing to do with parsing.
 SURFACES: dict[str, tuple[dict[str, Callable[[], Any]], Callable[[Any], Union[bytes, str]]]] = {
-    "db": (DB_BLOBS, lambda model: model.json()),
-    "backend_config": (BACKEND_STORED_CONFIGS, lambda model: model.json()),
-    "backend_creds": (BACKEND_CREDS, lambda model: model.json()),
-    "backend_data": (BACKEND_DATA, lambda model: model.json()),
-    "api_request": (API_REQUESTS, lambda model: model.json()),
-    "api_response": (API_RESPONSES, lambda model: bytes(CustomORJSONResponse(model).body)),
-    "runner": (RUNNER_REQUESTS, lambda model: model.json()),
-    "gateway": (GATEWAY_RESPONSES, lambda model: model.json()),
-    "proxy": (PROXY_REQUESTS, lambda model: json.dumps(model.dict(exclude_unset=True))),
-    "proxy_response": (PROXY_RESPONSES, lambda model: model.json()),
+    "db": (DB_BLOBS, lambda model: model.model_dump_json()),
+    "backend_config": (BACKEND_STORED_CONFIGS, lambda model: model.model_dump_json()),
+    "backend_creds": (BACKEND_CREDS, lambda model: model.model_dump_json()),
+    "backend_data": (BACKEND_DATA, lambda model: model.model_dump_json()),
+    "api_request": (API_REQUESTS, lambda model: model.model_dump_json()),
+    "api_response": (API_RESPONSES, lambda model: bytes(CustomJSONResponse(model).body)),
+    "runner": (
+        RUNNER_REQUESTS,
+        lambda model: model.json_for_runner()
+        if hasattr(model, "json_for_runner")
+        else model.model_dump_json(),
+    ),
+    "gateway": (GATEWAY_RESPONSES, lambda model: model.model_dump_json()),
+    "proxy": (PROXY_REQUESTS, lambda model: json.dumps(model.model_dump(exclude_unset=True))),
+    "proxy_response": (PROXY_RESPONSES, lambda model: model.model_dump_json()),
 }
 
 _CASES = [(surface, name) for surface, (reg, _) in SURFACES.items() for name in sorted(reg)]
@@ -160,7 +166,7 @@ class TestFleetNodesTargetCompatHack:
     """
     Pins the #3066 old-client hack explicitly, not just via the fixture bytes.
 
-    `FleetNodesSpec.dict()` drops `target` when it equals `min`. A fixture would catch the change
+    `FleetNodesSpec.model_dump()` drops `target` when it equals `min`. A fixture would catch the change
     but not explain it; naming the invariant gives the v2 `@model_serializer` rewrite something
     unambiguous to satisfy.
 
@@ -170,10 +176,10 @@ class TestFleetNodesTargetCompatHack:
     """
 
     def test_target_is_omitted_when_it_equals_min(self):
-        assert "target" not in FleetNodesSpec(min=1, target=1, max=1).dict()
+        assert "target" not in FleetNodesSpec(min=1, target=1, max=1).model_dump()
 
     def test_target_is_kept_when_it_differs_from_min(self):
-        assert FleetNodesSpec(min=1, target=5, max=5).dict()["target"] == 5
+        assert FleetNodesSpec(min=1, target=5, max=5).model_dump()["target"] == 5
 
     @pytest.mark.parametrize(
         ("nodes", "expected"),
@@ -191,8 +197,8 @@ def test_target_is_kept_when_it_differs_from_min(self):
         ],
     )
     def test_dict_and_json_apply_the_same_override(self, nodes, expected):
-        assert nodes.dict() == expected
-        assert json.loads(nodes.json()) == expected
+        assert nodes.model_dump() == expected
+        assert json.loads(nodes.model_dump_json()) == expected
 
     @pytest.mark.parametrize(
         ("nodes", "expected"),
@@ -212,8 +218,8 @@ def test_dict_and_json_apply_the_same_override(self, nodes, expected):
     def test_override_is_applied_when_nested(self, nodes, expected):
         configuration = FleetConfiguration(nodes=nodes)
 
-        assert configuration.dict()["nodes"] == expected
-        assert json.loads(configuration.json())["nodes"] == expected
+        assert configuration.model_dump()["nodes"] == expected
+        assert json.loads(configuration.model_dump_json())["nodes"] == expected
 
     def test_the_api_fixture_exercises_the_hack(self):
         nodes = factories.fleet().spec.configuration.nodes
@@ -243,10 +249,10 @@ def test_dict_and_json_apply_the_same_override(self, arch, expected):
             cpu=CPUSpec(arch=arch, count=Range[int](min=2, max=8)),
         )
 
-        assert json.loads(resources.json())["cpu"] == expected
+        assert json.loads(resources.model_dump_json())["cpu"] == expected
         # CoreModel.json() must call the overridden dict(); this assertion would expose a drift
         # even if only one of the two methods retained the compatibility rewrite.
-        assert canonicalize(json.dumps(resources.dict()["cpu"])) == canonicalize(
+        assert canonicalize(json.dumps(resources.model_dump()["cpu"])) == canonicalize(
             json.dumps(expected)
         )
 
@@ -269,15 +275,15 @@ def test_override_is_applied_when_nested(self, arch, expected):
             ),
         )
 
-        assert json.loads(configuration.json())["resources"]["cpu"] == expected
-        assert canonicalize(json.dumps(configuration.dict()["resources"]["cpu"])) == canonicalize(
-            json.dumps(expected)
-        )
+        assert json.loads(configuration.model_dump_json())["resources"]["cpu"] == expected
+        assert canonicalize(
+            json.dumps(configuration.model_dump()["resources"]["cpu"])
+        ) == canonicalize(json.dumps(expected))
 
 
 class TestFieldSerializationFilters:
     def test_submit_body_nested_field_includes_are_preserved(self):
-        body = factories.submit_body().dict()
+        body = json.loads(factories.submit_body().json_for_runner())
 
         assert set(body["run"]) == {"id", "run_spec"}
         assert set(body["run"]["run_spec"]) == {
@@ -309,15 +315,15 @@ def test_derived_merged_profile_is_excluded_from_dict_and_json(self):
         spec = factories.run_spec()
         assert spec.merged_profile is not None
 
-        assert "merged_profile" not in spec.dict()
-        assert "merged_profile" not in json.loads(spec.json())
+        assert "merged_profile" not in spec.model_dump()
+        assert "merged_profile" not in json.loads(spec.model_dump_json())
 
     def test_remote_repo_diff_bytes_are_excluded_from_dict_and_json(self):
         repo = RemoteRunRepoData(repo_name="dstack", repo_diff=b"secret diff")
         assert repo.repo_diff == b"secret diff"
 
-        assert "repo_diff" not in repo.dict()
-        assert "repo_diff" not in json.loads(repo.json())
+        assert "repo_diff" not in repo.model_dump()
+        assert "repo_diff" not in json.loads(repo.model_dump_json())
 
     def test_runpod_compatibility_availability_zone_is_excluded_directly_and_nested(self):
         configuration = RunpodVolumeConfiguration(
@@ -328,19 +334,19 @@ def test_runpod_compatibility_availability_zone_is_excluded_directly_and_nested(
         )
         assert configuration.availability_zone == "legacy-zone"
 
-        assert "availability_zone" not in configuration.dict()
-        assert "availability_zone" not in json.loads(configuration.json())
+        assert "availability_zone" not in configuration.model_dump()
+        assert "availability_zone" not in json.loads(configuration.model_dump_json())
 
         spec = VolumeSpec(configuration=configuration)
-        assert "availability_zone" not in spec.dict()["configuration"]
-        assert "availability_zone" not in json.loads(spec.json())["configuration"]
+        assert "availability_zone" not in spec.model_dump()["configuration"]
+        assert "availability_zone" not in json.loads(spec.model_dump_json())["configuration"]
 
 
-class TestCustomORJSONResponseCompat:
+class TestCustomJSONResponseCompat:
     def test_response_uses_the_same_nested_serializer_overrides_as_model_json(self):
         configuration = FleetConfiguration(nodes=FleetNodesSpec(min=1, target=1, max=1))
 
-        response_body = bytes(CustomORJSONResponse(configuration).body)
+        response_body = bytes(CustomJSONResponse(configuration).body)
 
-        assert canonicalize(response_body) == canonicalize(configuration.json())
+        assert canonicalize(response_body) == canonicalize(configuration.model_dump_json())
         assert "target" not in json.loads(response_body)["nodes"]
diff --git a/src/tests/_internal/server/background/pipeline_tasks/test_gateway_replicas.py b/src/tests/_internal/server/background/pipeline_tasks/test_gateway_replicas.py
index 6665d243df..e3af1c699a 100644
--- a/src/tests/_internal/server/background/pipeline_tasks/test_gateway_replicas.py
+++ b/src/tests/_internal/server/background/pipeline_tasks/test_gateway_replicas.py
@@ -95,7 +95,7 @@ async def test_fetch_selects_eligible_replicas_and_sets_lock_fields(
             region=None,
             status=GatewayReplicaStatus.SUBMITTED,
             last_processed_at=stale - timedelta(seconds=3),
-            configuration=get_gateway_compute_configuration().json(),
+            configuration=get_gateway_compute_configuration().model_dump_json(),
         )
         provisioning = await create_gateway_compute(
             session=session,
@@ -131,7 +131,7 @@ async def test_fetch_selects_eligible_replicas_and_sets_lock_fields(
             instance_id=None,
             region=None,
             last_processed_at=now,
-            configuration=get_gateway_compute_configuration().json(),
+            configuration=get_gateway_compute_configuration().model_dump_json(),
         )
         recent.created_at = now - timedelta(minutes=2)
         recent.last_processed_at = now
@@ -143,7 +143,7 @@ async def test_fetch_selects_eligible_replicas_and_sets_lock_fields(
             instance_id=None,
             region=None,
             last_processed_at=stale + timedelta(seconds=1),
-            configuration=get_gateway_compute_configuration().json(),
+            configuration=get_gateway_compute_configuration().model_dump_json(),
         )
         locked.lock_expires_at = now + timedelta(minutes=1)
         locked.lock_token = uuid.uuid4()
@@ -335,7 +335,7 @@ async def test_submitted_to_provisioning(
             instance_id=None,
             region=None,
             status=GatewayReplicaStatus.SUBMITTED,
-            configuration=get_gateway_compute_configuration().json(),
+            configuration=get_gateway_compute_configuration().model_dump_json(),
         )
         _lock_compute(compute)
         await session.commit()
@@ -379,7 +379,7 @@ async def test_submitted_backend_error_marks_terminated(
             instance_id=None,
             region=None,
             status=GatewayReplicaStatus.SUBMITTED,
-            configuration=get_gateway_compute_configuration().json(),
+            configuration=get_gateway_compute_configuration().model_dump_json(),
         )
         _lock_compute(compute)
         await session.commit()
@@ -417,7 +417,7 @@ async def test_submitted_backend_not_available_marks_terminated(
             instance_id=None,
             region=None,
             status=GatewayReplicaStatus.SUBMITTED,
-            configuration=get_gateway_compute_configuration().json(),
+            configuration=get_gateway_compute_configuration().model_dump_json(),
         )
         _lock_compute(compute)
         await session.commit()
@@ -453,7 +453,7 @@ async def test_submitted_skips_provisioning_if_gateway_to_be_deleted(
             instance_id=None,
             region=None,
             status=GatewayReplicaStatus.SUBMITTED,
-            configuration=get_gateway_compute_configuration().json(),
+            configuration=get_gateway_compute_configuration().model_dump_json(),
         )
         _lock_compute(compute)
         await session.commit()
@@ -488,7 +488,7 @@ async def test_submitted_skips_provisioning_if_gateway_failed(
             instance_id=None,
             region=None,
             status=GatewayReplicaStatus.SUBMITTED,
-            configuration=get_gateway_compute_configuration().json(),
+            configuration=get_gateway_compute_configuration().model_dump_json(),
         )
         _lock_compute(compute)
         await session.commit()
@@ -523,7 +523,7 @@ async def test_submitted_unexpected_error_marks_terminated(
             instance_id=None,
             region=None,
             status=GatewayReplicaStatus.SUBMITTED,
-            configuration=get_gateway_compute_configuration().json(),
+            configuration=get_gateway_compute_configuration().model_dump_json(),
         )
         _lock_compute(compute)
         await session.commit()
@@ -562,7 +562,7 @@ async def test_submitted_to_terminated_when_scaled_in(
             instance_id=None,
             region=None,
             status=GatewayReplicaStatus.SUBMITTED,
-            configuration=get_gateway_compute_configuration().json(),
+            configuration=get_gateway_compute_configuration().model_dump_json(),
         )
         compute.scale_in = True
         _lock_compute(compute)
diff --git a/src/tests/_internal/server/background/pipeline_tasks/test_gateways.py b/src/tests/_internal/server/background/pipeline_tasks/test_gateways.py
index 9de2f6acb2..e8c8b45a54 100644
--- a/src/tests/_internal/server/background/pipeline_tasks/test_gateways.py
+++ b/src/tests/_internal/server/background/pipeline_tasks/test_gateways.py
@@ -314,7 +314,7 @@ async def test_fetch_includes_running_gateway_with_pending_scale_attempt_even_if
             instance_id=None,
             region=None,
             status=GatewayReplicaStatus.SUBMITTED,
-            configuration=get_gateway_compute_configuration().json(),
+            configuration=get_gateway_compute_configuration().model_dump_json(),
         )
         gateway.replica_scale_attempt = 1
         await session.commit()
@@ -885,7 +885,7 @@ async def test_still_provisioning_with_submitted_replica(
             instance_id=None,
             region=None,
             status=GatewayReplicaStatus.SUBMITTED,
-            configuration=get_gateway_compute_configuration().json(),
+            configuration=get_gateway_compute_configuration().model_dump_json(),
         )
         gateway.lock_token = uuid.uuid4()
         gateway.lock_expires_at = datetime(2025, 1, 2, 3, 4, tzinfo=timezone.utc)
@@ -1237,7 +1237,7 @@ async def test_scale_in_prefers_less_advanced_replicas_over_older_running_ones(
             ip_address=None,
             instance_id=None,
             region=None,
-            configuration=get_gateway_compute_configuration().json(),
+            configuration=get_gateway_compute_configuration().model_dump_json(),
         )
         submitted.created_at = datetime(2025, 1, 2)
         gateway.lock_token = uuid.uuid4()
@@ -1425,7 +1425,7 @@ async def test_attempt_counter_not_reset_while_replacement_replica_still_provisi
             region=None,
             status=GatewayReplicaStatus.PROVISIONING,
             replica_num=0,
-            configuration=get_gateway_compute_configuration().json(),
+            configuration=get_gateway_compute_configuration().model_dump_json(),
         )
         gateway.replica_scale_attempt = 2
         gateway.lock_token = uuid.uuid4()
diff --git a/src/tests/_internal/server/background/pipeline_tasks/test_instances/test_check.py b/src/tests/_internal/server/background/pipeline_tasks/test_instances/test_check.py
index 90bb351f96..4ff57e0fd6 100644
--- a/src/tests/_internal/server/background/pipeline_tasks/test_instances/test_check.py
+++ b/src/tests/_internal/server/background/pipeline_tasks/test_instances/test_check.py
@@ -10,6 +10,7 @@
 from sqlalchemy import select
 from sqlalchemy.ext.asyncio import AsyncSession
 
+from dstack._internal.core.models.common import validate_json_extra_ignore
 from dstack._internal.core.models.fleets import FleetNodesSpec
 from dstack._internal.core.models.health import HealthStatus
 from dstack._internal.core.models.instances import (
@@ -201,7 +202,7 @@ async def test_check_shim_stores_gpu_driver(
 
         assert check_instance_inner_mock.call_args.kwargs["check_instance_info"]
         assert instance.job_provisioning_data is not None
-        jpd = JobProvisioningData.__response__.parse_raw(instance.job_provisioning_data)
+        jpd = validate_json_extra_ignore(JobProvisioningData, instance.job_provisioning_data)
         assert jpd.gpu_driver is not None
         assert jpd.gpu_driver.vendor == AcceleratorVendor.NVIDIA
         assert jpd.gpu_driver.version == "570.86.15"
@@ -481,7 +482,7 @@ async def test_check_shim_check_instance_health(
         res = await session.execute(select(InstanceHealthCheckModel))
         health_check = res.scalars().one()
         assert health_check.status == HealthStatus.WARNING
-        assert health_check.response == health_response.json()
+        assert health_check.response == health_response.model_dump_json()
 
 
 @pytest.mark.asyncio
@@ -1063,7 +1064,9 @@ def test_sets_new_or_changed_driver(self, current_version):
             gpu_driver=GpuDriverInfo(vendor=AcceleratorVendor.NVIDIA, version="570.86.15"),
         )
         assert "job_provisioning_data" in update_map
-        parsed = JobProvisioningData.__response__.parse_raw(update_map["job_provisioning_data"])
+        parsed = validate_json_extra_ignore(
+            JobProvisioningData, update_map["job_provisioning_data"]
+        )
         assert parsed.gpu_driver is not None
         assert parsed.gpu_driver.version == "570.86.15"
 
diff --git a/src/tests/_internal/server/background/pipeline_tasks/test_running_jobs.py b/src/tests/_internal/server/background/pipeline_tasks/test_running_jobs.py
index 737a6d4dcc..ffb7d18028 100644
--- a/src/tests/_internal/server/background/pipeline_tasks/test_running_jobs.py
+++ b/src/tests/_internal/server/background/pipeline_tasks/test_running_jobs.py
@@ -16,13 +16,14 @@
 from dstack._internal import settings
 from dstack._internal.core.errors import SSHError
 from dstack._internal.core.models.backends.base import BackendType
-from dstack._internal.core.models.common import NetworkMode
+from dstack._internal.core.models.common import NetworkMode, validate_json_extra_ignore
 from dstack._internal.core.models.configurations import (
     DevEnvironmentConfiguration,
     ProbeConfig,
     ServiceConfiguration,
     TaskConfiguration,
 )
+from dstack._internal.core.models.duration import Duration
 from dstack._internal.core.models.gateways import GatewayStatus
 from dstack._internal.core.models.instances import InstanceStatus
 from dstack._internal.core.models.profiles import StartupOrder, UtilizationPolicy
@@ -89,7 +90,7 @@
     get_volume_configuration,
     list_events,
 )
-from dstack._internal.utils.common import get_current_datetime
+from dstack._internal.utils.common import get_current_datetime, get_or_error
 
 pytestmark = pytest.mark.usefixtures("image_config_mock", "test_log_storage")
 
@@ -594,7 +595,9 @@ async def test_runs_provisioning_job(
         assert job.lock_expires_at is None
         assert job.lock_owner is None
         assert job.last_processed_at > before_processed_at
-        job_runtime_data = JobRuntimeData.__response__.parse_raw(job.job_runtime_data)
+        job_runtime_data = validate_json_extra_ignore(
+            JobRuntimeData, get_or_error(job.job_runtime_data)
+        )
         assert job_runtime_data.working_dir == "/dstack/run"
         assert job_runtime_data.username == "dstack"
 
@@ -829,7 +832,9 @@ async def test_pulling_shim(
         runner_client_mock.run_job.assert_called_once()
         await session.refresh(job)
         assert job.status == JobStatus.RUNNING
-        job_runtime_data = JobRuntimeData.__response__.parse_raw(job.job_runtime_data)
+        job_runtime_data = validate_json_extra_ignore(
+            JobRuntimeData, get_or_error(job.job_runtime_data)
+        )
         assert job_runtime_data.ports == {10022: 32771, 10999: 32772}
         assert job_runtime_data.working_dir == "/dstack/run"
         assert job_runtime_data.username == "dstack"
@@ -1227,7 +1232,9 @@ def assert_submit_job_to_runner(_, __, job_runtime_data, **kwargs):
 
         await session.refresh(job)
         assert job.status == JobStatus.PULLING
-        job_runtime_data = JobRuntimeData.__response__.parse_raw(job.job_runtime_data)
+        job_runtime_data = validate_json_extra_ignore(
+            JobRuntimeData, get_or_error(job.job_runtime_data)
+        )
         assert job_runtime_data.ports == expected_ports
 
     async def test_pulling_shim_failed(
@@ -1314,7 +1321,7 @@ async def test_pulling_shim_stores_pull_progress(
 
         await session.refresh(job)
         assert job.status == JobStatus.PULLING
-        assert job.image_pull_progress == progress.json()
+        assert job.image_pull_progress == progress.model_dump_json()
 
     async def test_provisioning_shim_force_stop_if_already_running_api_v1(
         self,
@@ -1800,7 +1807,7 @@ async def test_gpu_utilization(
                     ide="vscode",
                     utilization_policy=UtilizationPolicy(
                         min_gpu_utilization=80,
-                        time_window=600,
+                        time_window=Duration(600),
                     ),
                 ),
             ),
@@ -2453,7 +2460,7 @@ async def test_provisioning_shim_uses_server_default_registry(
 
 
 def _router_service_configuration(router_type: str) -> ServiceConfiguration:
-    return ServiceConfiguration.parse_obj(
+    return ServiceConfiguration.model_validate(
         {
             "type": "service",
             "port": 8000,
@@ -2510,7 +2517,7 @@ async def test_router_failed_terminates_worker(self):
             result.job_update_map.get("termination_reason_message") or ""
         )
 
-    @freeze_time("2023-01-01 12:00:00+00:00")
+    @freeze_time("2023-01-01 12:00:00Z")
     async def test_router_not_provisioned_within_timeout_defers(self):
         context = self._make_context(
             submitted_at=datetime(2023, 1, 1, 11, 45, 0, tzinfo=timezone.utc),
@@ -2524,7 +2531,7 @@ async def test_router_not_provisioned_within_timeout_defers(self):
         assert out is None
         assert result.job_update_map == {}
 
-    @freeze_time("2023-01-01 12:00:00+00:00")
+    @freeze_time("2023-01-01 12:00:00Z")
     async def test_router_not_provisioned_past_timeout_terminates(self):
         context = self._make_context(
             submitted_at=datetime(2023, 1, 1, 10, 0, 0, tzinfo=timezone.utc),
@@ -2627,7 +2634,7 @@ async def test_dynamo_run_loads_all_non_terminated_replicas(
             status=JobStatus.PROVISIONING,
         )
         run_id = run.id
-        parsed = RunSpec.__response__.parse_raw(run.run_spec)
+        parsed = validate_json_extra_ignore(RunSpec, get_or_error(run.run_spec))
         await session.commit()
         session.expire_all()
         run_model = await _fetch_run_model(
@@ -2663,7 +2670,7 @@ async def test_non_dynamo_loads_only_own_replica(self, test_db, session: AsyncSe
             status=JobStatus.PROVISIONING,
         )
         run_id = run.id
-        parsed = RunSpec.__response__.parse_raw(run.run_spec)
+        parsed = validate_json_extra_ignore(RunSpec, get_or_error(run.run_spec))
         await session.commit()
         session.expire_all()
         run_model = await _fetch_run_model(
diff --git a/src/tests/_internal/server/background/pipeline_tasks/test_runs/test_active.py b/src/tests/_internal/server/background/pipeline_tasks/test_runs/test_active.py
index 8c91fce0ba..470b7824d9 100644
--- a/src/tests/_internal/server/background/pipeline_tasks/test_runs/test_active.py
+++ b/src/tests/_internal/server/background/pipeline_tasks/test_runs/test_active.py
@@ -12,6 +12,7 @@
     ServiceConfiguration,
     TaskConfiguration,
 )
+from dstack._internal.core.models.duration import Duration
 from dstack._internal.core.models.instances import InstanceStatus
 from dstack._internal.core.models.profiles import (
     Profile,
@@ -185,7 +186,7 @@ async def test_retries_failed_replica_within_retry_duration(
             repo_id=repo.name,
             profile=Profile(
                 name="default",
-                retry=ProfileRetry(duration=3600, on_events=[RetryEvent.ERROR]),
+                retry=ProfileRetry(duration=Duration(3600), on_events=[RetryEvent.ERROR]),
             ),
         )
         run = await create_run(
@@ -227,7 +228,7 @@ async def test_retries_no_capacity_replica_and_keeps_service_running(
             repo_id=repo.name,
             profile=Profile(
                 name="default",
-                retry=ProfileRetry(duration=3600, on_events=[RetryEvent.INTERRUPTION]),
+                retry=ProfileRetry(duration=Duration(3600), on_events=[RetryEvent.INTERRUPTION]),
             ),
             configuration=ServiceConfiguration(
                 port=8080,
@@ -306,7 +307,7 @@ async def test_replica_retry_deletes_superseded_no_capacity_submissions(
             repo_id=repo.name,
             profile=Profile(
                 name="default",
-                retry=ProfileRetry(duration=3600, on_events=[RetryEvent.INTERRUPTION]),
+                retry=ProfileRetry(duration=Duration(3600), on_events=[RetryEvent.INTERRUPTION]),
             ),
             configuration=ServiceConfiguration(
                 port=8080,
@@ -386,7 +387,7 @@ async def test_retries_scheduled_run_no_capacity_from_trigger_time(
             repo_id=repo.name,
             profile=Profile(
                 name="default",
-                retry=ProfileRetry(duration=3600, on_events=[RetryEvent.NO_CAPACITY]),
+                retry=ProfileRetry(duration=Duration(3600), on_events=[RetryEvent.NO_CAPACITY]),
             ),
             configuration=TaskConfiguration(
                 commands=["echo hello"],
@@ -435,7 +436,7 @@ async def test_terminates_scheduled_run_when_no_capacity_retry_exceeded_from_tri
             repo_id=repo.name,
             profile=Profile(
                 name="default",
-                retry=ProfileRetry(duration=600, on_events=[RetryEvent.NO_CAPACITY]),
+                retry=ProfileRetry(duration=Duration(600), on_events=[RetryEvent.NO_CAPACITY]),
             ),
             configuration=TaskConfiguration(
                 commands=["echo hello"],
@@ -484,7 +485,7 @@ async def test_retrying_multinode_replica_terminates_active_sibling_jobs(
             repo_id=repo.name,
             profile=Profile(
                 name="default",
-                retry=ProfileRetry(duration=3600, on_events=[RetryEvent.ERROR]),
+                retry=ProfileRetry(duration=Duration(3600), on_events=[RetryEvent.ERROR]),
             ),
             configuration=TaskConfiguration(
                 commands=["echo hello"],
@@ -548,7 +549,7 @@ async def test_transitions_to_pending_when_retry_duration_exceeded(
             repo_id=repo.name,
             profile=Profile(
                 name="default",
-                retry=ProfileRetry(duration=60, on_events=[RetryEvent.ERROR]),
+                retry=ProfileRetry(duration=Duration(60), on_events=[RetryEvent.ERROR]),
             ),
         )
         run = await create_run(
@@ -984,7 +985,7 @@ async def test_service_rolling_deployment_scale_up(
         # cannot be applied and rolling deployment is triggered instead.
         old_spec = get_job_spec(old_job)
         old_spec.commands = ["echo old!"]
-        old_job.job_spec_data = old_spec.json()
+        old_job.job_spec_data = old_spec.model_dump_json()
         await session.commit()
 
         lock_run(run)
@@ -1053,7 +1054,7 @@ async def test_service_rolling_deployment_scale_down_old_unregistered(
         )
         old_spec = get_job_spec(old_job)
         old_spec.commands = ["echo old!"]
-        old_job.job_spec_data = old_spec.json()
+        old_job.job_spec_data = old_spec.model_dump_json()
         await session.commit()
 
         lock_run(run)
@@ -1111,7 +1112,7 @@ async def test_service_removed_group_cleanup(
         # Patch the job spec to have replica_group="old"
         old_spec = get_job_spec(old_group_job)
         old_spec.replica_group = "old"
-        old_group_job.job_spec_data = old_spec.json()
+        old_group_job.job_spec_data = old_spec.model_dump_json()
         await session.commit()
 
         lock_run(run)
diff --git a/src/tests/_internal/server/background/pipeline_tasks/test_submitted_jobs.py b/src/tests/_internal/server/background/pipeline_tasks/test_submitted_jobs.py
index 30c038945d..ee34ec5ad5 100644
--- a/src/tests/_internal/server/background/pipeline_tasks/test_submitted_jobs.py
+++ b/src/tests/_internal/server/background/pipeline_tasks/test_submitted_jobs.py
@@ -11,7 +11,12 @@
 
 from dstack._internal.core.errors import BackendError
 from dstack._internal.core.models.backends.base import BackendType
-from dstack._internal.core.models.common import EntityReference, NetworkMode, RegistryAuth
+from dstack._internal.core.models.common import (
+    EntityReference,
+    NetworkMode,
+    RegistryAuth,
+    validate_json_extra_ignore,
+)
 from dstack._internal.core.models.configurations import ServiceConfiguration, TaskConfiguration
 from dstack._internal.core.models.envs import Env
 from dstack._internal.core.models.fleets import FleetNodesSpec, InstanceGroupPlacement
@@ -74,7 +79,7 @@
     get_ssh_fleet_configuration,
     get_volume_provisioning_data,
 )
-from dstack._internal.utils.common import get_current_datetime
+from dstack._internal.utils.common import get_current_datetime, get_or_error
 
 pytestmark = pytest.mark.usefixtures("image_config_mock")
 
@@ -1548,8 +1553,12 @@ async def test_assigns_multinode_jobs_to_specific_shared_ssh_instances(
         assert worker_job.instance is not None and worker_job.instance.id == selected_worker.id
         assert selected_master.busy_blocks == 2
         assert selected_worker.busy_blocks == 2
-        master_runtime = JobRuntimeData.__response__.parse_raw(master_job.job_runtime_data)
-        worker_runtime = JobRuntimeData.__response__.parse_raw(worker_job.job_runtime_data)
+        master_runtime = validate_json_extra_ignore(
+            JobRuntimeData, get_or_error(master_job.job_runtime_data)
+        )
+        worker_runtime = validate_json_extra_ignore(
+            JobRuntimeData, get_or_error(worker_job.job_runtime_data)
+        )
         assert master_runtime.network_mode == NetworkMode.HOST
         assert worker_runtime.network_mode == NetworkMode.HOST
         assert master_runtime.offer is not None and master_runtime.offer.blocks == 2
@@ -2566,7 +2575,7 @@ async def test_interpolates_secrets_when_provisioning_new_capacity(
             repo_id=repo.name,
             configuration=TaskConfiguration(
                 image="ubuntu",
-                env=Env.parse_obj({"TOKEN": "${{ secrets.token }}"}),
+                env=Env.model_validate({"TOKEN": "${{ secrets.token }}"}),
                 registry_auth=RegistryAuth(
                     username="${{ secrets.registry_user }}",
                     password="${{ secrets.registry_pass }}",
diff --git a/src/tests/_internal/server/background/scheduled_tasks/test_idle_volumes.py b/src/tests/_internal/server/background/scheduled_tasks/test_idle_volumes.py
index 86aeea0f4c..e2db9f4254 100644
--- a/src/tests/_internal/server/background/scheduled_tasks/test_idle_volumes.py
+++ b/src/tests/_internal/server/background/scheduled_tasks/test_idle_volumes.py
@@ -173,8 +173,7 @@ async def test_volume_attached(self, test_db, session: AsyncSession):
         project = await create_project(session=session)
         user = await create_user(session=session)
 
-        config = get_volume_configuration(name="test-volume")
-        config.auto_cleanup_duration = "1h"
+        config = get_volume_configuration(name="test-volume", auto_cleanup_duration=3600)
 
         volume = await create_volume(
             session=session,
@@ -198,8 +197,7 @@ async def test_idle_duration_threshold(self, test_db, session: AsyncSession):
         project = await create_project(session=session)
         user = await create_user(session=session)
 
-        config = get_volume_configuration(name="test-volume")
-        config.auto_cleanup_duration = "1h"
+        config = get_volume_configuration(name="test-volume", auto_cleanup_duration=3600)
 
         volume = await create_volume(
             session=session,
diff --git a/src/tests/_internal/server/conftest.py b/src/tests/_internal/server/conftest.py
index 3894ac3601..125cc5de17 100644
--- a/src/tests/_internal/server/conftest.py
+++ b/src/tests/_internal/server/conftest.py
@@ -17,6 +17,33 @@
 )
 
 
+def _warm_up_route_schemas() -> None:
+    """
+    Build every route's pydantic schemas once, at import time.
+
+    FastAPI builds a route's dependant lazily, on the first request that matches it. Several tests
+    make that first request inside `@freeze_time`, where `datetime.datetime` is freezegun's
+    `FakeDatetime`. pydantic v2 matches `datetime` by exact type and rejects the subclass, so a
+    route with an `Optional[datetime]` query parameter (`after`, `before`) fails to build with
+    `PydanticSchemaGenerationError` — and because `FakeDatetime` mimics `datetime`'s repr, the
+    error names `datetime.datetime` and reads like a production bug.
+
+    Doing this eagerly, before any test freezes the clock, keeps the schemas built from the real
+    types. Nothing in production freezes time, so this is a test-environment fix only.
+    """
+    # FastAPI 0.141 defers this to `_IncludedRouter.effective_candidates()`, reached from
+    # `matches()` on the first request. Duck-typed rather than importing the private class.
+    pending = list(app.routes)
+    while pending:
+        route = pending.pop()
+        build = getattr(route, "effective_candidates", None)
+        if callable(build):
+            pending.extend(build())
+
+
+_warm_up_route_schemas()
+
+
 @pytest.fixture
 def client():
     transport = httpx.ASGITransport(app=app)
@@ -34,7 +61,9 @@ def test_log_storage(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> FileLog
 
 @pytest.fixture
 def image_config_mock(monkeypatch: pytest.MonkeyPatch) -> ImageConfig:
-    image_config = ImageConfig.parse_obj({"User": None, "Entrypoint": None, "Cmd": ["/bin/bash"]})
+    image_config = ImageConfig.model_validate(
+        {"User": None, "Entrypoint": None, "Cmd": ["/bin/bash"]}
+    )
     monkeypatch.setattr(
         "dstack._internal.server.services.jobs.configurators.base._get_image_config",
         Mock(return_value=image_config),
diff --git a/src/tests/_internal/server/routers/test_events.py b/src/tests/_internal/server/routers/test_events.py
index 7dd6359af1..eed7015ec7 100644
--- a/src/tests/_internal/server/routers/test_events.py
+++ b/src/tests/_internal/server/routers/test_events.py
@@ -71,7 +71,7 @@ async def test_response_format(self, session: AsyncSession, client: AsyncClient)
             {
                 "id": str(event_ids[1]),
                 "message": "Project updated",
-                "recorded_at": "2026-01-01T12:00:01+00:00",
+                "recorded_at": "2026-01-01T12:00:01Z",
                 "actor_user_id": None,
                 "actor_user": None,
                 "is_actor_user_deleted": None,
@@ -89,7 +89,7 @@ async def test_response_format(self, session: AsyncSession, client: AsyncClient)
             {
                 "id": str(event_ids[0]),
                 "message": "User added to project",
-                "recorded_at": "2026-01-01T12:00:00+00:00",
+                "recorded_at": "2026-01-01T12:00:00Z",
                 "actor_user_id": str(user.id),
                 "actor_user": "test_user",
                 "is_actor_user_deleted": False,
diff --git a/src/tests/_internal/server/routers/test_fleets.py b/src/tests/_internal/server/routers/test_fleets.py
index b7a48b54df..c8f5507796 100644
--- a/src/tests/_internal/server/routers/test_fleets.py
+++ b/src/tests/_internal/server/routers/test_fleets.py
@@ -392,7 +392,7 @@ async def test_lists_fleets(self, test_db, session: AsyncSession, client: AsyncC
                 "name": fleet.name,
                 "project_name": project.name,
                 "spec": json.loads(fleet.spec),
-                "created_at": "2023-01-02T03:04:00+00:00",
+                "created_at": "2023-01-02T03:04:00Z",
                 "status": fleet.status.value,
                 "status_message": None,
                 "instances": [],
@@ -628,7 +628,7 @@ async def test_returns_fleet_by_id(
             "name": fleet.name,
             "project_name": project.name,
             "spec": json.loads(fleet.spec),
-            "created_at": "2023-01-02T03:04:00+00:00",
+            "created_at": "2023-01-02T03:04:00Z",
             "status": fleet.status.value,
             "status_message": None,
             "instances": [],
@@ -670,7 +670,7 @@ async def test_returns_not_deleted_fleet_by_name(
             "name": active_fleet.name,
             "project_name": project.name,
             "spec": json.loads(active_fleet.spec),
-            "created_at": "2023-01-02T03:04:00+00:00",
+            "created_at": "2023-01-02T03:04:00Z",
             "status": active_fleet.status.value,
             "status_message": None,
             "instances": [],
@@ -923,7 +923,7 @@ async def test_creates_fleet(self, test_db, session: AsyncSession, client: Async
         response = await client.post(
             f"/api/project/{project.name}/fleets/apply",
             headers=get_auth_headers(user.token),
-            json={"plan": {"spec": spec.dict()}, "force": False},
+            json={"plan": {"spec": spec.model_dump()}, "force": False},
         )
         assert response.status_code == 200
         assert response.json() == {
@@ -979,7 +979,7 @@ async def test_creates_fleet(self, test_db, session: AsyncSession, client: Async
                 },
                 "autocreated": False,
             },
-            "created_at": "2023-01-02T03:04:00+00:00",
+            "created_at": "2023-01-02T03:04:00Z",
             "status": "active",
             "status_message": None,
             "instances": [
@@ -997,7 +997,7 @@ async def test_creates_fleet(self, test_db, session: AsyncSession, client: Async
                     "health_status": "healthy",
                     "termination_reason": None,
                     "termination_reason_message": None,
-                    "created": "2023-01-02T03:04:00+00:00",
+                    "created": "2023-01-02T03:04:00Z",
                     "finished_at": None,
                     "backend": None,
                     "region": None,
@@ -1037,7 +1037,7 @@ async def test_creates_ssh_fleet(self, test_db, session: AsyncSession, client: A
         response = await client.post(
             f"/api/project/{project.name}/fleets/apply",
             headers=get_auth_headers(user.token),
-            json={"plan": {"spec": spec.dict()}, "force": False},
+            json={"plan": {"spec": spec.model_dump()}, "force": False},
         )
         assert response.status_code == 200, response.json()
         assert response.json() == {
@@ -1101,7 +1101,7 @@ async def test_creates_ssh_fleet(self, test_db, session: AsyncSession, client: A
                 },
                 "autocreated": False,
             },
-            "created_at": "2023-01-02T03:04:00+00:00",
+            "created_at": "2023-01-02T03:04:00Z",
             "status": "active",
             "status_message": None,
             "instances": [
@@ -1132,7 +1132,7 @@ async def test_creates_ssh_fleet(self, test_db, session: AsyncSession, client: A
                     "health_status": "healthy",
                     "termination_reason": None,
                     "termination_reason_message": None,
-                    "created": "2023-01-02T03:04:00+00:00",
+                    "created": "2023-01-02T03:04:00Z",
                     "finished_at": None,
                     "region": "remote",
                     "availability_zone": None,
@@ -1192,7 +1192,7 @@ async def test_creates_ssh_fleet_with_blocks(
         response = await client.post(
             f"/api/project/{project.name}/fleets/apply",
             headers=get_auth_headers(user.token),
-            json={"plan": {"spec": spec.dict()}, "force": False},
+            json={"plan": {"spec": spec.model_dump()}, "force": False},
         )
         assert response.status_code == 200, response.json()
         res = await session.execute(select(FleetModel))
@@ -1218,7 +1218,7 @@ async def test_updates_ssh_fleet(self, test_db, session: AsyncSession, client: A
             network=None,
         )
         current_spec = get_fleet_spec(conf=current_conf)
-        spec = current_spec.copy(deep=True)
+        spec = current_spec.model_copy(deep=True)
         # 10.0.0.100 removed, 10.0.0.101 added
         spec.configuration.ssh_config.hosts = ["10.0.0.101"]
 
@@ -1253,7 +1253,7 @@ async def test_updates_ssh_fleet(self, test_db, session: AsyncSession, client: A
             headers=get_auth_headers(user.token),
             json={
                 "plan": {
-                    "spec": spec.dict(),
+                    "spec": spec.model_dump(),
                     "current_resource": _fleet_model_to_json_dict(fleet),
                 },
                 "force": False,
@@ -1322,7 +1322,7 @@ async def test_updates_ssh_fleet(self, test_db, session: AsyncSession, client: A
                 },
                 "autocreated": False,
             },
-            "created_at": "2023-01-02T03:04:00+00:00",
+            "created_at": "2023-01-02T03:04:00Z",
             "status": "active",
             "status_message": None,
             "instances": [
@@ -1353,7 +1353,7 @@ async def test_updates_ssh_fleet(self, test_db, session: AsyncSession, client: A
                     "health_status": "healthy",
                     "termination_reason": "terminated_by_user",
                     "termination_reason_message": None,
-                    "created": "2023-01-02T03:04:00+00:00",
+                    "created": "2023-01-02T03:04:00Z",
                     "finished_at": None,
                     "region": "remote",
                     "availability_zone": None,
@@ -1389,7 +1389,7 @@ async def test_updates_ssh_fleet(self, test_db, session: AsyncSession, client: A
                     "health_status": "healthy",
                     "termination_reason": None,
                     "termination_reason_message": None,
-                    "created": "2023-01-02T03:04:00+00:00",
+                    "created": "2023-01-02T03:04:00Z",
                     "finished_at": None,
                     "region": "remote",
                     "availability_zone": None,
@@ -1436,7 +1436,7 @@ async def test_updates_cloud_fleet_nodes_in_place_when_fleet_in_use(
             status=InstanceStatus.BUSY,
             instance_num=0,
         )
-        spec = current_spec.copy(deep=True)
+        spec = current_spec.model_copy(deep=True)
         spec.configuration.nodes = FleetNodesSpec(min=1, target=1, max=3)
 
         response = await client.post(
@@ -1444,7 +1444,7 @@ async def test_updates_cloud_fleet_nodes_in_place_when_fleet_in_use(
             headers=get_auth_headers(user.token),
             json={
                 "plan": {
-                    "spec": spec.dict(),
+                    "spec": spec.model_dump(),
                     "current_resource": _fleet_model_to_json_dict(fleet),
                 },
                 "force": False,
@@ -1475,7 +1475,7 @@ async def test_updates_cloud_fleet_nodes_target_without_changing_instance_count(
             conf=get_fleet_configuration(nodes=FleetNodesSpec(min=0, target=0, max=1))
         )
         fleet = await create_fleet(session=session, project=project, spec=current_spec)
-        spec = current_spec.copy(deep=True)
+        spec = current_spec.model_copy(deep=True)
         spec.configuration.nodes = FleetNodesSpec(min=0, target=1, max=1)
 
         response = await client.post(
@@ -1483,7 +1483,7 @@ async def test_updates_cloud_fleet_nodes_target_without_changing_instance_count(
             headers=get_auth_headers(user.token),
             json={
                 "plan": {
-                    "spec": spec.dict(),
+                    "spec": spec.model_dump(),
                     "current_resource": _fleet_model_to_json_dict(fleet),
                 },
                 "force": False,
@@ -1538,7 +1538,7 @@ async def test_errors_if_ssh_key_is_bad(
         response = await client.post(
             f"/api/project/{project.name}/fleets/apply",
             headers=get_auth_headers(user.token),
-            json={"plan": {"spec": spec.dict()}, "force": False},
+            json={"plan": {"spec": spec.model_dump()}, "force": False},
         )
         assert response.status_code == 400
 
@@ -1573,7 +1573,7 @@ async def test_errors_if_ssh_fleet_uses_backend_only_field(
         response = await client.post(
             f"/api/project/{project.name}/fleets/apply",
             headers=get_auth_headers(user.token),
-            json={"plan": {"spec": spec.dict()}, "force": False},
+            json={"plan": {"spec": spec.model_dump()}, "force": False},
         )
         assert response.status_code == 400, response.json()
         assert response.json()["detail"][0]["msg"] == (
@@ -1583,7 +1583,7 @@ async def test_errors_if_ssh_fleet_uses_backend_only_field(
     @pytest.mark.parametrize(
         ["field_name", "field_value"],
         [
-            pytest.param("env", Env.parse_obj({"K": "V"}), id="env"),
+            pytest.param("env", Env.model_validate({"K": "V"}), id="env"),
         ],
     )
     @pytest.mark.asyncio
@@ -1607,7 +1607,7 @@ async def test_errors_if_backend_fleet_uses_ssh_only_field(
         response = await client.post(
             f"/api/project/{project.name}/fleets/apply",
             headers=get_auth_headers(user.token),
-            json={"plan": {"spec": spec.dict()}, "force": False},
+            json={"plan": {"spec": spec.model_dump()}, "force": False},
         )
         assert response.status_code == 400, response.json()
         assert response.json()["detail"][0]["msg"] == (
@@ -1641,7 +1641,7 @@ async def test_forbids_if_no_permission_to_manage_ssh_fleets(
             response = await client.post(
                 f"/api/project/{project.name}/fleets/apply",
                 headers=get_auth_headers(user.token),
-                json={"plan": {"spec": spec.dict()}, "force": False},
+                json={"plan": {"spec": spec.model_dump()}, "force": False},
             )
         assert response.status_code in [401, 403]
 
@@ -1678,7 +1678,7 @@ async def test_importer_member_cannot_apply_plan_on_imported_fleet(
         response = await client.post(
             f"/api/project/{exporter_project.name}/fleets/apply",
             headers=get_auth_headers(importer_user.token),
-            json={"plan": {"spec": spec.dict()}, "force": False},
+            json={"plan": {"spec": spec.model_dump()}, "force": False},
         )
         assert response.status_code == 403
 
@@ -2224,7 +2224,7 @@ async def test_returns_create_plan_for_new_fleet(
             response = await client.post(
                 f"/api/project/{project.name}/fleets/get_plan",
                 headers=get_auth_headers(user.token),
-                json={"spec": spec.dict()},
+                json={"spec": spec.model_dump()},
             )
             backend_mock.compute.return_value.get_offers.assert_called_once()
 
@@ -2232,10 +2232,10 @@ async def test_returns_create_plan_for_new_fleet(
         assert response.json() == {
             "project_name": project.name,
             "user": user.name,
-            "spec": json.loads(spec.json()),
-            "effective_spec": json.loads(spec.json()),
+            "spec": json.loads(spec.model_dump_json()),
+            "effective_spec": json.loads(spec.model_dump_json()),
             "current_resource": None,
-            "offers": [json.loads(o.json()) for o in offers],
+            "offers": [json.loads(o.model_dump_json()) for o in offers],
             "total_offers": len(offers),
             "max_offer_price": 1.0,
             "action": "create",
@@ -2267,13 +2267,13 @@ async def test_returns_offers_for_elastic_container_backend_fleet(
             response = await client.post(
                 f"/api/project/{project.name}/fleets/get_plan",
                 headers=get_auth_headers(user.token),
-                json={"spec": spec.dict()},
+                json={"spec": spec.model_dump()},
             )
             backend_mock.compute.return_value.get_offers.assert_called_once()
 
         response_json = response.json()
         assert response.status_code == 200, response_json
-        assert response_json["offers"] == [json.loads(offer.json())]
+        assert response_json["offers"] == [json.loads(offer.model_dump_json())]
         assert response_json["total_offers"] == 1
         assert response_json["max_offer_price"] == offer.price
 
@@ -2303,7 +2303,7 @@ async def test_returns_no_offers_for_non_elastic_container_backend_fleet(
             response = await client.post(
                 f"/api/project/{project.name}/fleets/get_plan",
                 headers=get_auth_headers(user.token),
-                json={"spec": spec.dict()},
+                json={"spec": spec.model_dump()},
             )
             backend_mock.compute.return_value.get_offers.assert_called_once()
 
@@ -2325,9 +2325,9 @@ async def test_returns_update_plan_for_existing_fleet(
         )
         conf = get_ssh_fleet_configuration(hosts=["10.0.0.100"])
         spec = get_fleet_spec(conf=conf)
-        effective_spec = spec.copy(deep=True)
+        effective_spec = spec.model_copy(deep=True)
         effective_spec.configuration.ssh_config.ssh_key = None
-        current_spec = spec.copy(deep=True)
+        current_spec = spec.model_copy(deep=True)
         # `hosts` can be updated in-place
         current_spec.configuration.ssh_config.hosts = ["10.0.0.100", "10.0.0.101"]
         fleet = await create_fleet(session=session, project=project, spec=current_spec)
@@ -2335,15 +2335,15 @@ async def test_returns_update_plan_for_existing_fleet(
         response = await client.post(
             f"/api/project/{project.name}/fleets/get_plan",
             headers=get_auth_headers(user.token),
-            json={"spec": spec.dict()},
+            json={"spec": spec.model_dump()},
         )
 
         assert response.status_code == 200
         assert response.json() == {
             "project_name": project.name,
             "user": user.name,
-            "spec": spec.dict(),
-            "effective_spec": effective_spec.dict(),
+            "spec": spec.model_dump(),
+            "effective_spec": effective_spec.model_dump(),
             "current_resource": _fleet_model_to_json_dict(fleet),
             "offers": [],
             "total_offers": 0,
@@ -2364,14 +2364,14 @@ async def test_returns_update_plan_for_existing_cloud_fleet_nodes_update(
         current_spec = get_fleet_spec(
             conf=get_fleet_configuration(nodes=FleetNodesSpec(min=0, target=0, max=1))
         )
-        spec = current_spec.copy(deep=True)
+        spec = current_spec.model_copy(deep=True)
         spec.configuration.nodes = FleetNodesSpec(min=1, target=1, max=1)
         fleet = await create_fleet(session=session, project=project, spec=current_spec)
 
         response = await client.post(
             f"/api/project/{project.name}/fleets/get_plan",
             headers=get_auth_headers(user.token),
-            json={"spec": spec.dict()},
+            json={"spec": spec.model_dump()},
         )
 
         response_json = response.json()
@@ -2392,14 +2392,14 @@ async def test_returns_create_plan_for_existing_cloud_fleet_blocks_update(
         current_spec = get_fleet_spec(
             conf=get_fleet_configuration(nodes=FleetNodesSpec(min=0, target=0, max=1))
         )
-        spec = current_spec.copy(deep=True)
+        spec = current_spec.model_copy(deep=True)
         spec.configuration.blocks = 2
         fleet = await create_fleet(session=session, project=project, spec=current_spec)
 
         response = await client.post(
             f"/api/project/{project.name}/fleets/get_plan",
             headers=get_auth_headers(user.token),
-            json={"spec": spec.dict()},
+            json={"spec": spec.model_dump()},
         )
 
         response_json = response.json()
@@ -2420,7 +2420,7 @@ async def test_returns_update_plan_for_existing_cloud_fleet_provisioning_fields_
         current_spec = get_fleet_spec(
             conf=get_fleet_configuration(nodes=FleetNodesSpec(min=0, target=0, max=1))
         )
-        spec = current_spec.copy(deep=True)
+        spec = current_spec.model_copy(deep=True)
         spec.configuration.backends = [BackendType.AWS]
         spec.configuration.regions = ["us-east-1"]
         fleet = await create_fleet(session=session, project=project, spec=current_spec)
@@ -2428,7 +2428,7 @@ async def test_returns_update_plan_for_existing_cloud_fleet_provisioning_fields_
         response = await client.post(
             f"/api/project/{project.name}/fleets/get_plan",
             headers=get_auth_headers(user.token),
-            json={"spec": spec.dict()},
+            json={"spec": spec.model_dump()},
         )
 
         response_json = response.json()
@@ -2448,9 +2448,9 @@ async def test_returns_create_plan_for_existing_fleet(
         )
         conf = get_ssh_fleet_configuration(placement=InstanceGroupPlacement.ANY)
         spec = get_fleet_spec(conf=conf)
-        effective_spec = spec.copy(deep=True)
+        effective_spec = spec.model_copy(deep=True)
         effective_spec.configuration.ssh_config.ssh_key = None
-        current_spec = spec.copy(deep=True)
+        current_spec = spec.model_copy(deep=True)
         # `placement` cannot be updated in-place
         current_spec.configuration.placement = InstanceGroupPlacement.CLUSTER
         fleet = await create_fleet(session=session, project=project, spec=current_spec)
@@ -2458,15 +2458,15 @@ async def test_returns_create_plan_for_existing_fleet(
         response = await client.post(
             f"/api/project/{project.name}/fleets/get_plan",
             headers=get_auth_headers(user.token),
-            json={"spec": spec.dict()},
+            json={"spec": spec.model_dump()},
         )
 
         assert response.status_code == 200
         assert response.json() == {
             "project_name": project.name,
             "user": user.name,
-            "spec": spec.dict(),
-            "effective_spec": effective_spec.dict(),
+            "spec": spec.model_dump(),
+            "effective_spec": effective_spec.model_dump(),
             "current_resource": _fleet_model_to_json_dict(fleet),
             "offers": [],
             "total_offers": 0,
@@ -2527,7 +2527,7 @@ async def test_replaces_no_balance_with_not_available_for_old_clients(
             response = await client.post(
                 f"/api/project/{project.name}/fleets/get_plan",
                 headers=headers,
-                json={"spec": get_fleet_spec().dict()},
+                json={"spec": get_fleet_spec().model_dump()},
             )
 
         assert response.status_code == 200
@@ -2565,10 +2565,10 @@ async def test_importer_member_cannot_get_plan_for_imported_fleet(
         response = await client.post(
             f"/api/project/{exporter_project.name}/fleets/get_plan",
             headers=get_auth_headers(importer_user.token),
-            json={"spec": spec.dict()},
+            json={"spec": spec.model_dump()},
         )
         assert response.status_code == 403
 
 
 def _fleet_model_to_json_dict(fleet: FleetModel) -> dict:
-    return json.loads(fleet_model_to_fleet(fleet).json())
+    return json.loads(fleet_model_to_fleet(fleet).model_dump_json())
diff --git a/src/tests/_internal/server/routers/test_gpus.py b/src/tests/_internal/server/routers/test_gpus.py
index 18cfee03db..85b38257e5 100644
--- a/src/tests/_internal/server/routers/test_gpus.py
+++ b/src/tests/_internal/server/routers/test_gpus.py
@@ -142,7 +142,7 @@ async def call_gpus_api(
     unallocated_resources: Optional[bool] = None,
 ):
     """Helper to call the GPUs API with standard parameters."""
-    json_data = {"run_spec": run_spec.dict()}
+    json_data = {"run_spec": run_spec.model_dump()}
     if group_by is not None:
         json_data["group_by"] = group_by
     if full_offers is not None:
@@ -439,7 +439,7 @@ async def test_returns_empty_gpus_when_no_offers(
             response = await client.post(
                 f"/api/project/{project.name}/gpus/list",
                 headers=get_auth_headers(user.token),
-                json={"run_spec": run_spec.dict()},
+                json={"run_spec": run_spec.model_dump()},
             )
 
         assert response.status_code == 200
@@ -465,7 +465,7 @@ async def test_invalid_group_by_rejected(
         response = await client.post(
             f"/api/project/{project.name}/gpus/list",
             headers=get_auth_headers(user.token),
-            json={"run_spec": run_spec.dict(), "group_by": ["invalid_field"]},
+            json={"run_spec": run_spec.model_dump(), "group_by": ["invalid_field"]},
         )
         assert response.status_code == 422
         assert "validation error" in response.text.lower() or "invalid" in response.text.lower()
@@ -602,7 +602,7 @@ async def test_exact_aggregation_values(
             response = await client.post(
                 f"/api/project/{project.name}/gpus/list",
                 headers=get_auth_headers(user.token),
-                json={"run_spec": run_spec.dict()},
+                json={"run_spec": run_spec.model_dump()},
             )
             assert response.status_code == 200
             data = response.json()
@@ -626,7 +626,7 @@ async def test_exact_aggregation_values(
             response_count_grouped = await client.post(
                 f"/api/project/{project.name}/gpus/list",
                 headers=get_auth_headers(user.token),
-                json={"run_spec": run_spec.dict(), "group_by": ["count"]},
+                json={"run_spec": run_spec.model_dump(), "group_by": ["count"]},
             )
             assert response_count_grouped.status_code == 200
             count_grouped_data = response_count_grouped.json()
@@ -666,7 +666,7 @@ async def test_exact_aggregation_values(
             response_backend = await client.post(
                 f"/api/project/{project.name}/gpus/list",
                 headers=get_auth_headers(user.token),
-                json={"run_spec": run_spec.dict(), "group_by": ["backend"]},
+                json={"run_spec": run_spec.model_dump(), "group_by": ["backend"]},
             )
             assert response_backend.status_code == 200
             backend_data = response_backend.json()
@@ -710,7 +710,7 @@ async def test_exact_aggregation_values(
             response_region = await client.post(
                 f"/api/project/{project.name}/gpus/list",
                 headers=get_auth_headers(user.token),
-                json={"run_spec": run_spec.dict(), "group_by": ["backend", "region"]},
+                json={"run_spec": run_spec.model_dump(), "group_by": ["backend", "region"]},
             )
             assert response_region.status_code == 200
             region_data = response_region.json()
diff --git a/src/tests/_internal/server/routers/test_instances.py b/src/tests/_internal/server/routers/test_instances.py
index 439538c14c..43f0eb1c44 100644
--- a/src/tests/_internal/server/routers/test_instances.py
+++ b/src/tests/_internal/server/routers/test_instances.py
@@ -596,17 +596,17 @@ async def test_returns_health_checks(self, session: AsyncSession, client: AsyncC
         assert response.json() == {
             "health_checks": [
                 {
-                    "collected_at": "2025-01-01T12:01:00+00:00",
+                    "collected_at": "2025-01-01T12:01:00Z",
                     "status": "failure",
                     "events": [
                         {
-                            "timestamp": "2025-01-01T12:01:00+00:00",
+                            "timestamp": "2025-01-01T12:01:00Z",
                             "status": "failure",
                             "message": "Detected 333 volatile double-bit ECC error(s) in GPU 0.",
                         }
                     ],
                 },
-                {"collected_at": "2025-01-01T12:00:00+00:00", "status": "healthy", "events": []},
+                {"collected_at": "2025-01-01T12:00:00Z", "status": "healthy", "events": []},
             ]
         }
 
diff --git a/src/tests/_internal/server/routers/test_logs.py b/src/tests/_internal/server/routers/test_logs.py
index 33d904d56a..6c0d5e056c 100644
--- a/src/tests/_internal/server/routers/test_logs.py
+++ b/src/tests/_internal/server/routers/test_logs.py
@@ -43,9 +43,9 @@ async def test_returns_logs(
         )
         runner_log_path.parent.mkdir(parents=True, exist_ok=True)
         runner_log_path.write_text(
-            '{"timestamp": "2023-10-06T10:01:53.234234+00:00", "log_source": "stdout", "message": "Hello"}\n'
-            '{"timestamp": "2023-10-06T10:01:53.234235+00:00", "log_source": "stdout", "message": "World"}\n'
-            '{"timestamp": "2023-10-06T10:01:53.234236+00:00", "log_source": "stdout", "message": "!"}\n'
+            '{"timestamp": "2023-10-06T10:01:53.234234Z", "log_source": "stdout", "message": "Hello"}\n'
+            '{"timestamp": "2023-10-06T10:01:53.234235Z", "log_source": "stdout", "message": "World"}\n'
+            '{"timestamp": "2023-10-06T10:01:53.234236Z", "log_source": "stdout", "message": "!"}\n'
         )
         response = await client.post(
             f"/api/project/{project.name}/logs/poll",
@@ -60,17 +60,17 @@ async def test_returns_logs(
         assert response.json() == {
             "logs": [
                 {
-                    "timestamp": "2023-10-06T10:01:53.234234+00:00",
+                    "timestamp": "2023-10-06T10:01:53.234234Z",
                     "log_source": "stdout",
                     "message": "SGVsbG8=",
                 },
                 {
-                    "timestamp": "2023-10-06T10:01:53.234235+00:00",
+                    "timestamp": "2023-10-06T10:01:53.234235Z",
                     "log_source": "stdout",
                     "message": "V29ybGQ=",
                 },
                 {
-                    "timestamp": "2023-10-06T10:01:53.234236+00:00",
+                    "timestamp": "2023-10-06T10:01:53.234236Z",
                     "log_source": "stdout",
                     "message": "IQ==",
                 },
@@ -84,7 +84,7 @@ async def test_returns_logs(
             json={
                 "run_name": "test_run",
                 "job_submission_id": "1b0e1b45-2f8c-4ab6-8010-a0d1a3e44e0e",
-                "start_time": "2023-10-06T10:01:53.234235+00:00",
+                "start_time": "2023-10-06T10:01:53.234235Z",
                 "diagnose": True,
             },
         )
@@ -92,7 +92,7 @@ async def test_returns_logs(
         assert response.json() == {
             "logs": [
                 {
-                    "timestamp": "2023-10-06T10:01:53.234236+00:00",
+                    "timestamp": "2023-10-06T10:01:53.234236Z",
                     "log_source": "stdout",
                     "message": "IQ==",
                 },
diff --git a/src/tests/_internal/server/routers/test_metrics.py b/src/tests/_internal/server/routers/test_metrics.py
index 19e849f413..d747c7ade7 100644
--- a/src/tests/_internal/server/routers/test_metrics.py
+++ b/src/tests/_internal/server/routers/test_metrics.py
@@ -107,47 +107,47 @@ async def test_returns_metrics(self, test_db, session: AsyncSession, client: Asy
             "metrics": [
                 {
                     "name": "cpu_usage_percent",
-                    "timestamps": ["2023-01-02T03:04:25+00:00"],
+                    "timestamps": ["2023-01-02T03:04:25Z"],
                     "values": [60],
                 },
                 {
                     "name": "memory_usage_bytes",
-                    "timestamps": ["2023-01-02T03:04:25+00:00"],
+                    "timestamps": ["2023-01-02T03:04:25Z"],
                     "values": [1024],
                 },
                 {
                     "name": "memory_working_set_bytes",
-                    "timestamps": ["2023-01-02T03:04:25+00:00"],
+                    "timestamps": ["2023-01-02T03:04:25Z"],
                     "values": [512],
                 },
                 {
                     "name": "cpus_detected_num",
-                    "timestamps": ["2023-01-02T03:04:25+00:00"],
+                    "timestamps": ["2023-01-02T03:04:25Z"],
                     "values": [64],
                 },
                 {
                     "name": "memory_total_bytes",
-                    "timestamps": ["2023-01-02T03:04:25+00:00"],
+                    "timestamps": ["2023-01-02T03:04:25Z"],
                     "values": [137438953472],
                 },
                 {
                     "name": "gpus_detected_num",
-                    "timestamps": ["2023-01-02T03:04:25+00:00"],
+                    "timestamps": ["2023-01-02T03:04:25Z"],
                     "values": [1],
                 },
                 {
                     "name": "gpu_memory_total_bytes",
-                    "timestamps": ["2023-01-02T03:04:25+00:00"],
+                    "timestamps": ["2023-01-02T03:04:25Z"],
                     "values": [34359738368],
                 },
                 {
                     "name": "gpu_memory_usage_bytes_gpu0",
-                    "timestamps": ["2023-01-02T03:04:25+00:00"],
+                    "timestamps": ["2023-01-02T03:04:25Z"],
                     "values": [1024],
                 },
                 {
                     "name": "gpu_util_percent_gpu0",
-                    "timestamps": ["2023-01-02T03:04:25+00:00"],
+                    "timestamps": ["2023-01-02T03:04:25Z"],
                     "values": [10],
                 },
             ]
diff --git a/src/tests/_internal/server/routers/test_projects.py b/src/tests/_internal/server/routers/test_projects.py
index 67afa77390..b15fabfd6f 100644
--- a/src/tests/_internal/server/routers/test_projects.py
+++ b/src/tests/_internal/server/routers/test_projects.py
@@ -65,7 +65,7 @@ async def test_returns_projects(self, test_db, session: AsyncSession, client: As
                 "owner": {
                     "id": str(user.id),
                     "username": user.name,
-                    "created_at": "2023-01-02T03:04:00+00:00",
+                    "created_at": "2023-01-02T03:04:00Z",
                     "global_role": user.global_role,
                     "email": None,
                     "active": True,
@@ -74,7 +74,7 @@ async def test_returns_projects(self, test_db, session: AsyncSession, client: As
                     },
                     "ssh_public_key": None,
                 },
-                "created_at": "2023-01-02T03:04:00+00:00",
+                "created_at": "2023-01-02T03:04:00Z",
                 "backends": [],
                 "members": [],
                 "is_public": False,
@@ -245,7 +245,7 @@ async def test_returns_paginated_projects(
                 "owner": {
                     "id": str(user.id),
                     "username": user.name,
-                    "created_at": "2023-01-02T03:00:00+00:00",
+                    "created_at": "2023-01-02T03:00:00Z",
                     "global_role": user.global_role,
                     "email": None,
                     "active": True,
@@ -254,7 +254,7 @@ async def test_returns_paginated_projects(
                     },
                     "ssh_public_key": None,
                 },
-                "created_at": "2023-01-02T03:06:00+00:00",
+                "created_at": "2023-01-02T03:06:00Z",
                 "backends": [],
                 "members": [],
                 "is_public": False,
@@ -265,7 +265,7 @@ async def test_returns_paginated_projects(
             "/api/projects/list",
             headers=get_auth_headers(user.token),
             json={
-                "prev_created_at": "2023-01-02T03:06:00+00:00",
+                "prev_created_at": "2023-01-02T03:06:00Z",
                 "prev_id": str(project3.id),
                 "limit": 1,
             },
@@ -278,7 +278,7 @@ async def test_returns_paginated_projects(
                 "owner": {
                     "id": str(user.id),
                     "username": user.name,
-                    "created_at": "2023-01-02T03:00:00+00:00",
+                    "created_at": "2023-01-02T03:00:00Z",
                     "global_role": user.global_role,
                     "email": None,
                     "active": True,
@@ -287,7 +287,7 @@ async def test_returns_paginated_projects(
                     },
                     "ssh_public_key": None,
                 },
-                "created_at": "2023-01-02T03:05:00+00:00",
+                "created_at": "2023-01-02T03:05:00Z",
                 "backends": [],
                 "members": [],
                 "is_public": False,
@@ -298,7 +298,7 @@ async def test_returns_paginated_projects(
             "/api/projects/list",
             headers=get_auth_headers(user.token),
             json={
-                "prev_created_at": "2023-01-02T03:05:00+00:00",
+                "prev_created_at": "2023-01-02T03:05:00Z",
                 "prev_id": str(project2.id),
                 "limit": 1,
             },
@@ -311,7 +311,7 @@ async def test_returns_paginated_projects(
                 "owner": {
                     "id": str(user.id),
                     "username": user.name,
-                    "created_at": "2023-01-02T03:00:00+00:00",
+                    "created_at": "2023-01-02T03:00:00Z",
                     "global_role": user.global_role,
                     "email": None,
                     "active": True,
@@ -320,7 +320,7 @@ async def test_returns_paginated_projects(
                     },
                     "ssh_public_key": None,
                 },
-                "created_at": "2023-01-02T03:04:00+00:00",
+                "created_at": "2023-01-02T03:04:00Z",
                 "backends": [],
                 "members": [],
                 "is_public": False,
@@ -363,7 +363,7 @@ async def test_returns_total_count(self, test_db, session: AsyncSession, client:
                     "owner": {
                         "id": str(user.id),
                         "username": user.name,
-                        "created_at": "2023-01-02T03:00:00+00:00",
+                        "created_at": "2023-01-02T03:00:00Z",
                         "global_role": user.global_role,
                         "email": None,
                         "active": True,
@@ -372,7 +372,7 @@ async def test_returns_total_count(self, test_db, session: AsyncSession, client:
                         },
                         "ssh_public_key": None,
                     },
-                    "created_at": "2023-01-02T03:05:00+00:00",
+                    "created_at": "2023-01-02T03:05:00Z",
                     "backends": [],
                     "members": [],
                     "is_public": False,
@@ -948,7 +948,7 @@ async def test_creates_project(self, test_db, session: AsyncSession, client: Asy
             "owner": {
                 "id": str(user.id),
                 "username": user.name,
-                "created_at": "2023-01-02T03:04:00+00:00",
+                "created_at": "2023-01-02T03:04:00Z",
                 "global_role": user.global_role,
                 "email": None,
                 "active": True,
@@ -957,14 +957,14 @@ async def test_creates_project(self, test_db, session: AsyncSession, client: Asy
                 },
                 "ssh_public_key": user.ssh_public_key,
             },
-            "created_at": "2023-01-02T03:04:00+00:00",
+            "created_at": "2023-01-02T03:04:00Z",
             "backends": [],
             "members": [
                 {
                     "user": {
                         "id": str(user.id),
                         "username": user.name,
-                        "created_at": "2023-01-02T03:04:00+00:00",
+                        "created_at": "2023-01-02T03:04:00Z",
                         "global_role": user.global_role,
                         "email": None,
                         "active": True,
@@ -1509,7 +1509,7 @@ async def test_returns_project(self, test_db, session: AsyncSession, client: Asy
             "owner": {
                 "id": str(user.id),
                 "username": user.name,
-                "created_at": "2023-01-02T03:04:00+00:00",
+                "created_at": "2023-01-02T03:04:00Z",
                 "global_role": user.global_role,
                 "email": None,
                 "active": True,
@@ -1518,14 +1518,14 @@ async def test_returns_project(self, test_db, session: AsyncSession, client: Asy
                 },
                 "ssh_public_key": None,
             },
-            "created_at": "2023-01-02T03:04:00+00:00",
+            "created_at": "2023-01-02T03:04:00Z",
             "backends": [],
             "members": [
                 {
                     "user": {
                         "id": str(user.id),
                         "username": user.name,
-                        "created_at": "2023-01-02T03:04:00+00:00",
+                        "created_at": "2023-01-02T03:04:00Z",
                         "global_role": user.global_role,
                         "email": None,
                         "active": True,
@@ -1756,7 +1756,7 @@ async def test_sets_project_members(self, test_db, session: AsyncSession, client
                 "user": {
                     "id": str(admin.id),
                     "username": admin.name,
-                    "created_at": "2023-01-02T03:04:00+00:00",
+                    "created_at": "2023-01-02T03:04:00Z",
                     "global_role": admin.global_role,
                     "email": None,
                     "active": True,
@@ -1775,7 +1775,7 @@ async def test_sets_project_members(self, test_db, session: AsyncSession, client
                 "user": {
                     "id": str(user1.id),
                     "username": user1.name,
-                    "created_at": "2023-01-02T03:04:00+00:00",
+                    "created_at": "2023-01-02T03:04:00Z",
                     "global_role": user1.global_role,
                     "email": None,
                     "active": True,
@@ -1794,7 +1794,7 @@ async def test_sets_project_members(self, test_db, session: AsyncSession, client
                 "user": {
                     "id": str(user2.id),
                     "username": user2.name,
-                    "created_at": "2023-01-02T03:04:00+00:00",
+                    "created_at": "2023-01-02T03:04:00Z",
                     "global_role": user2.global_role,
                     "email": None,
                     "active": True,
@@ -1852,7 +1852,7 @@ async def test_sets_project_members_by_email(
                 "user": {
                     "id": str(user1.id),
                     "username": user1.name,
-                    "created_at": "2023-01-02T03:04:00+00:00",
+                    "created_at": "2023-01-02T03:04:00Z",
                     "global_role": user1.global_role,
                     "email": user1.email,
                     "active": True,
diff --git a/src/tests/_internal/server/routers/test_public_keys.py b/src/tests/_internal/server/routers/test_public_keys.py
index 80954e0262..e4dee3895b 100644
--- a/src/tests/_internal/server/routers/test_public_keys.py
+++ b/src/tests/_internal/server/routers/test_public_keys.py
@@ -43,7 +43,7 @@ async def test_lists_own_public_keys(self, session: AsyncSession, client: AsyncC
         assert response.json() == [
             {
                 "id": str(key.id),
-                "added_at": "2023-01-02T03:04:00+00:00",
+                "added_at": "2023-01-02T03:04:00Z",
                 "name": "my-key",
                 "type": "ssh-ed25519",
                 "fingerprint": "SHA256:testfingerprint",
@@ -137,7 +137,7 @@ async def test_adds_valid_public_key(
             "type": "ssh-ed25519",
             "name": "test@example.com",
             "fingerprint": "SHA256:uALbfMqe7g4MMaRS5NMJen38dAEHwtxzR0iX0Ymuc80",
-            "added_at": "2023-01-02T03:04:00+00:00",
+            "added_at": "2023-01-02T03:04:00Z",
         }
         validate_openssh_public_key_mock.assert_awaited_once_with(self.PUBLIC_KEY)
 
diff --git a/src/tests/_internal/server/routers/test_runs.py b/src/tests/_internal/server/routers/test_runs.py
index 0eeb4a11ac..46d530fff1 100644
--- a/src/tests/_internal/server/routers/test_runs.py
+++ b/src/tests/_internal/server/routers/test_runs.py
@@ -79,6 +79,7 @@
     list_events,
 )
 from dstack._internal.server.testing.matchers import SomeUUID4Str
+from dstack._internal.utils.common import render_datetime_as_api
 
 pytestmark = pytest.mark.usefixtures("image_config_mock", "disable_sshproxy")
 
@@ -187,7 +188,7 @@ def get_dev_env_run_plan_dict(
                 "gpu": None,
                 "shm_size": None,
             },
-            "volumes": [json.loads(v.json()) for v in volumes],
+            "volumes": [json.loads(v.model_dump_json()) for v in volumes],
             "repos": [
                 {
                     "url": "https://github.com/dstackai/dstack",
@@ -303,7 +304,7 @@ def get_dev_env_run_plan_dict(
                         "backend_options": None,
                     },
                     "retry": None,
-                    "volumes": volumes,
+                    "volumes": [json.loads(v.model_dump_json()) for v in volumes],
                     "ssh_key": None,
                     "working_dir": None,
                     "repo_code_hash": None,
@@ -321,12 +322,12 @@ def get_dev_env_run_plan_dict(
                     "service_port": None,
                     "probes": [],
                 },
-                "offers": [json.loads(o.json()) for o in offers],
+                "offers": [json.loads(o.model_dump_json()) for o in offers],
                 "total_offers": total_offers,
                 "max_price": max_price,
             }
         ],
-        "current_resource": current_resource.dict() if current_resource else None,
+        "current_resource": current_resource.model_dump() if current_resource else None,
         "action": action.value,
     }
 
@@ -338,9 +339,9 @@ def get_dev_env_run_dict(
     username: str = "test_user",
     run_name: Optional[str] = "run_name",
     repo_id: str = "test_repo",
-    submitted_at: str = "2023-01-02T03:04:00+00:00",
-    last_processed_at: str = "2023-01-02T03:04:00+00:00",
-    finished_at: Optional[str] = "2023-01-02T03:04:00+00:00",
+    submitted_at: str = "2023-01-02T03:04:00Z",
+    last_processed_at: str = "2023-01-02T03:04:00Z",
+    finished_at: Optional[str] = "2023-01-02T03:04:00Z",
     privileged: bool = False,
     docker: Optional[bool] = None,
     deleted: bool = False,
@@ -695,14 +696,14 @@ async def test_lists_runs(self, test_db, session: AsyncSession, client: AsyncCli
             fleet=fleet,
             submitted_at=run1_submitted_at,
         )
-        run1_spec = RunSpec.parse_raw(run1.run_spec)
+        run1_spec = RunSpec.model_validate_json(run1.run_spec)
         job = await create_job(
             session=session,
             run=run1,
             submitted_at=run1_submitted_at,
             last_processed_at=run1_submitted_at,
         )
-        job_spec = JobSpec.parse_raw(job.job_spec_data)
+        job_spec = JobSpec.model_validate_json(job.job_spec_data)
         run2_submitted_at = datetime(2023, 1, 1, 3, 4, tzinfo=timezone.utc)
         run2 = await create_run(
             session=session,
@@ -712,7 +713,7 @@ async def test_lists_runs(self, test_db, session: AsyncSession, client: AsyncCli
             fleet=fleet,
             submitted_at=run2_submitted_at,
         )
-        run2_spec = RunSpec.parse_raw(run2.run_spec)
+        run2_spec = RunSpec.model_validate_json(run2.run_spec)
         response = await client.post(
             "/api/runs/list",
             headers=get_auth_headers(user.token),
@@ -728,21 +729,21 @@ async def test_lists_runs(self, test_db, session: AsyncSession, client: AsyncCli
                     "id": str(fleet.id),
                     "name": fleet.name,
                 },
-                "submitted_at": run1_submitted_at.isoformat(),
-                "last_processed_at": run1_submitted_at.isoformat(),
+                "submitted_at": render_datetime_as_api(run1_submitted_at),
+                "last_processed_at": render_datetime_as_api(run1_submitted_at),
                 "status": "submitted",
                 "status_message": "submitted",
-                "run_spec": run1_spec.dict(),
+                "run_spec": run1_spec.model_dump(),
                 "jobs": [
                     {
-                        "job_spec": job_spec.dict(),
+                        "job_spec": job_spec.model_dump(),
                         "job_submissions": [
                             {
                                 "id": str(job.id),
                                 "submission_num": 0,
                                 "deployment_num": 0,
-                                "submitted_at": run1_submitted_at.isoformat(),
-                                "last_processed_at": run1_submitted_at.isoformat(),
+                                "submitted_at": render_datetime_as_api(run1_submitted_at),
+                                "last_processed_at": render_datetime_as_api(run1_submitted_at),
                                 "finished_at": None,
                                 "inactivity_secs": None,
                                 "status": "submitted",
@@ -764,8 +765,8 @@ async def test_lists_runs(self, test_db, session: AsyncSession, client: AsyncCli
                     "id": str(job.id),
                     "submission_num": 0,
                     "deployment_num": 0,
-                    "submitted_at": run1_submitted_at.isoformat(),
-                    "last_processed_at": run1_submitted_at.isoformat(),
+                    "submitted_at": render_datetime_as_api(run1_submitted_at),
+                    "last_processed_at": render_datetime_as_api(run1_submitted_at),
                     "finished_at": None,
                     "inactivity_secs": None,
                     "status": "submitted",
@@ -779,7 +780,7 @@ async def test_lists_runs(self, test_db, session: AsyncSession, client: AsyncCli
                     "probes": [],
                     "image_pull_progress": None,
                 },
-                "cost": 0,
+                "cost": 0.0,
                 "service": None,
                 "deployment_num": 0,
                 "termination_reason": None,
@@ -795,14 +796,14 @@ async def test_lists_runs(self, test_db, session: AsyncSession, client: AsyncCli
                     "id": str(fleet.id),
                     "name": fleet.name,
                 },
-                "submitted_at": run2_submitted_at.isoformat(),
-                "last_processed_at": run2_submitted_at.isoformat(),
+                "submitted_at": render_datetime_as_api(run2_submitted_at),
+                "last_processed_at": render_datetime_as_api(run2_submitted_at),
                 "status": "submitted",
                 "status_message": "submitted",
-                "run_spec": run2_spec.dict(),
+                "run_spec": run2_spec.model_dump(),
                 "jobs": [],
                 "latest_job_submission": None,
-                "cost": 0,
+                "cost": 0.0,
                 "service": None,
                 "deployment_num": 0,
                 "termination_reason": None,
@@ -895,7 +896,7 @@ async def test_limits_job_submissions(
             user=user,
             submitted_at=run_submitted_at,
         )
-        run_spec = RunSpec.parse_raw(run.run_spec)
+        run_spec = RunSpec.model_validate_json(run.run_spec)
         await create_job(
             session=session,
             run=run,
@@ -909,7 +910,7 @@ async def test_limits_job_submissions(
             submitted_at=run_submitted_at,
             last_processed_at=run_submitted_at,
         )
-        job2_spec = JobSpec.parse_raw(job2.job_spec_data)
+        job2_spec = JobSpec.model_validate_json(job2.job_spec_data)
         response = await client.post(
             "/api/runs/list",
             headers=get_auth_headers(user.token),
@@ -922,21 +923,21 @@ async def test_limits_job_submissions(
                 "project_name": project.name,
                 "user": user.name,
                 "fleet": None,
-                "submitted_at": run_submitted_at.isoformat(),
-                "last_processed_at": run_submitted_at.isoformat(),
+                "submitted_at": render_datetime_as_api(run_submitted_at),
+                "last_processed_at": render_datetime_as_api(run_submitted_at),
                 "status": "submitted",
                 "status_message": "submitted",
-                "run_spec": run_spec.dict(),
+                "run_spec": run_spec.model_dump(),
                 "jobs": [
                     {
-                        "job_spec": job2_spec.dict(),
+                        "job_spec": job2_spec.model_dump(),
                         "job_submissions": [
                             {
                                 "id": str(job2.id),
                                 "submission_num": 1,
                                 "deployment_num": 0,
-                                "submitted_at": run_submitted_at.isoformat(),
-                                "last_processed_at": run_submitted_at.isoformat(),
+                                "submitted_at": render_datetime_as_api(run_submitted_at),
+                                "last_processed_at": render_datetime_as_api(run_submitted_at),
                                 "finished_at": None,
                                 "inactivity_secs": None,
                                 "status": "submitted",
@@ -958,8 +959,8 @@ async def test_limits_job_submissions(
                     "id": str(job2.id),
                     "submission_num": 1,
                     "deployment_num": 0,
-                    "submitted_at": run_submitted_at.isoformat(),
-                    "last_processed_at": run_submitted_at.isoformat(),
+                    "submitted_at": render_datetime_as_api(run_submitted_at),
+                    "last_processed_at": render_datetime_as_api(run_submitted_at),
                     "finished_at": None,
                     "inactivity_secs": None,
                     "status": "submitted",
@@ -973,7 +974,7 @@ async def test_limits_job_submissions(
                     "probes": [],
                     "image_pull_progress": None,
                 },
-                "cost": 0,
+                "cost": 0.0,
                 "service": None,
                 "deployment_num": 0,
                 "termination_reason": None,
@@ -1603,7 +1604,7 @@ async def test_forwards_full_offers_to_compute_get_offers(
             repo_id=repo.name,
             configuration=DevEnvironmentConfiguration(ide="vscode"),
         )
-        body: dict = {"run_spec": json.loads(run_spec.json())}
+        body: dict = {"run_spec": json.loads(run_spec.model_dump_json())}
         if body_full_offers is not None:
             body["full_offers"] = body_full_offers
 
@@ -1658,7 +1659,7 @@ async def test_forwards_unallocated_resources_to_compute_get_offers(
             repo_id=repo.name,
             configuration=DevEnvironmentConfiguration(ide="vscode"),
         )
-        body: dict = {"run_spec": json.loads(run_spec.json())}
+        body: dict = {"run_spec": json.loads(run_spec.model_dump_json())}
         if body_unallocated_resources is not None:
             body["unallocated_resources"] = body_unallocated_resources
 
@@ -1774,7 +1775,7 @@ async def test_task_with_two_nodes_returns_two_job_plans(
             repo_id=repo.name,
             configuration=TaskConfiguration(commands=["echo hi"], nodes=2),
         )
-        body = {"run_spec": json.loads(run_spec.json())}
+        body = {"run_spec": json.loads(run_spec.model_dump_json())}
         with patch("dstack._internal.server.services.backends.get_project_backends") as m:
             backend_mock = Mock()
             backend_mock.TYPE = BackendType.AWS
@@ -1853,7 +1854,7 @@ async def test_service_with_two_replica_groups_returns_two_job_plans(
                 ],
             ),
         )
-        body = {"run_spec": json.loads(run_spec.json())}
+        body = {"run_spec": json.loads(run_spec.model_dump_json())}
 
         def offers_by_requirements(
             requirements: Requirements, full_offers: bool, unallocated_resources: bool
@@ -1914,7 +1915,7 @@ async def test_service_reservation_group_filters_backends_by_reservation_support
                 ],
             ),
         )
-        body = {"run_spec": json.loads(run_spec.json())}
+        body = {"run_spec": json.loads(run_spec.model_dump_json())}
 
         with patch("dstack._internal.server.services.backends.get_project_backends") as m:
             aws_backend_mock = Mock()
@@ -2371,7 +2372,7 @@ async def test_preserves_backend_specific_offer_order(
         run_spec = get_run_spec(
             repo_id=repo.name, configuration=parse_run_configuration(configuration)
         )
-        body = {"run_spec": run_spec.dict()}
+        body = {"run_spec": run_spec.model_dump()}
 
         backend_mock_aws = Mock()
         backend_mock_aws.TYPE = BackendType.AWS
@@ -2442,7 +2443,7 @@ async def test_offer_cli_preserves_backend_specific_offer_order_across_fleets(
                 fleets=["fleet-aws", "fleet-vastai"],
             ),
         )
-        body = {"run_spec": run_spec.dict()}
+        body = {"run_spec": run_spec.model_dump()}
 
         backend_mock_aws = Mock()
         backend_mock_aws.TYPE = BackendType.AWS
@@ -2543,7 +2544,7 @@ async def test_offer_cli_returns_offers_from_all_specified_fleets(
         response = await client.post(
             f"/api/project/{project.name}/runs/get_plan",
             headers=get_auth_headers(user.token),
-            json={"run_spec": run_spec.dict()},
+            json={"run_spec": run_spec.model_dump()},
         )
 
         assert response.status_code == 200, response.json()
@@ -2590,7 +2591,7 @@ async def test_offer_cli_deduplicates_identical_backend_offers_across_specified_
                 fleets=["fleet-a", "fleet-b"],
             ),
         )
-        body = {"run_spec": run_spec.dict()}
+        body = {"run_spec": run_spec.model_dump()}
 
         with patch("dstack._internal.server.services.backends.get_project_backends") as m:
             backend_mock_aws = Mock()
@@ -2678,7 +2679,7 @@ async def test_offer_cli_keeps_identical_existing_instances_from_specified_fleet
         response = await client.post(
             f"/api/project/{project.name}/runs/get_plan",
             headers=get_auth_headers(user.token),
-            json={"run_spec": run_spec.dict()},
+            json={"run_spec": run_spec.model_dump()},
         )
 
         assert response.status_code == 200, response.json()
@@ -2712,7 +2713,7 @@ async def test_offer_cli_without_fleet_keeps_global_offers(
                 user="root",
             ),
         )
-        body = {"run_spec": run_spec.dict()}
+        body = {"run_spec": run_spec.model_dump()}
         with patch("dstack._internal.server.services.backends.get_project_backends") as m:
             backend_mock_aws = Mock()
             backend_mock_aws.TYPE = BackendType.AWS
@@ -2804,7 +2805,7 @@ async def test_offer_without_fleets_uses_global_offer_collection(
             response = await client.post(
                 f"/api/project/{project.name}/runs/get_plan",
                 headers=get_auth_headers(user.token),
-                json={"run_spec": run_spec.dict()},
+                json={"run_spec": run_spec.model_dump()},
             )
 
         assert response.status_code == 200, response.json()
@@ -2867,7 +2868,7 @@ async def test_offer_with_fleets_uses_selected_fleet_offer_collection(
             response = await client.post(
                 f"/api/project/{project.name}/runs/get_plan",
                 headers=get_auth_headers(user.token),
-                json={"run_spec": run_spec.dict()},
+                json={"run_spec": run_spec.model_dump()},
             )
 
         assert response.status_code == 200, response.json()
@@ -2931,7 +2932,7 @@ async def test_regular_run_plan_uses_best_fleet_candidate_selection(
             response = await client.post(
                 f"/api/project/{project.name}/runs/get_plan",
                 headers=get_auth_headers(user.token),
-                json={"run_spec": run_spec.dict()},
+                json={"run_spec": run_spec.model_dump()},
             )
 
         assert response.status_code == 200, response.json()
@@ -3110,12 +3111,12 @@ async def test_returns_update_or_create_action_on_conf_change(
         response = await client.post(
             f"/api/project/{project.name}/runs/get_plan",
             headers=get_auth_headers(user.token),
-            json={"run_spec": run_spec.dict()},
+            json={"run_spec": run_spec.model_dump()},
         )
         assert response.status_code == 200
         response_json = response.json()
         assert response_json["action"] == action
-        assert response_json["current_resource"] == json.loads(run.json())
+        assert response_json["current_resource"] == json.loads(run.model_dump_json())
 
     @pytest.mark.asyncio
     @pytest.mark.usefixtures("test_db")
@@ -3133,7 +3134,7 @@ async def test_generates_user_ssh_key(self, session: AsyncSession, client: Async
         response = await client.post(
             f"/api/project/{project.name}/runs/get_plan",
             headers=get_auth_headers(user.token),
-            json={"run_spec": run_spec.dict()},
+            json={"run_spec": run_spec.model_dump()},
         )
 
         assert response.status_code == 200, response.json()
@@ -3184,7 +3185,7 @@ async def test_patches_service_configuration_probes_for_old_clients(
             run_name="test-service",
         )
 
-        body = {"run_spec": run_spec.dict()}
+        body = {"run_spec": run_spec.model_dump()}
         headers = get_auth_headers(user.token)
         if client_version is not None:
             headers["X-API-Version"] = client_version
@@ -3228,7 +3229,7 @@ async def test_submits_new_run_if_no_current_resource(
             session=session, project=project, user=user, project_role=ProjectRole.USER
         )
         submitted_at = datetime(2023, 1, 2, 3, 4, tzinfo=timezone.utc)
-        submitted_at_formatted = "2023-01-02T03:04:00+00:00"
+        submitted_at_formatted = "2023-01-02T03:04:00Z"
         last_processed_at_formatted = submitted_at_formatted
         repo = await create_repo(session=session, project_id=project.id)
         run_dict = get_dev_env_run_dict(
@@ -3296,7 +3297,7 @@ async def test_updates_run(self, test_db, session: AsyncSession, client: AsyncCl
         )
         run = run_model_to_run(run_model)
         run_spec.configuration_path = "new.dstack.yml"
-        run_spec.configuration.replicas = Range(min=2, max=2)
+        run_spec.configuration.replicas = Range[int](min=2, max=2)
         response = await client.post(
             f"/api/project/{project.name}/runs/apply",
             headers=get_auth_headers(user.token),
@@ -3308,7 +3309,7 @@ async def test_updates_run(self, test_db, session: AsyncSession, client: AsyncCl
                         current_resource=run,
                     ),
                     force=False,
-                ).json()
+                ).model_dump_json()
             ),
         )
         assert response.status_code == 200, response.json()
@@ -3348,7 +3349,7 @@ async def test_creates_pending_run_if_run_is_scheduled(
                 headers=get_auth_headers(user.token),
                 json={
                     "plan": {
-                        "run_spec": json.loads(run_spec.json()),
+                        "run_spec": json.loads(run_spec.model_dump_json()),
                         "current_resource": None,
                     },
                     "force": False,
@@ -3379,7 +3380,7 @@ async def test_generates_user_ssh_key(self, session: AsyncSession, client: Async
             headers=get_auth_headers(user.token),
             json={
                 "plan": {
-                    "run_spec": run_spec.dict(),
+                    "run_spec": run_spec.model_dump(),
                     "current_resource": None,
                 },
                 "force": False,
@@ -3434,7 +3435,7 @@ async def test_patches_service_configuration_probes_for_old_clients(
             headers=headers,
             json={
                 "plan": {
-                    "run_spec": run_spec.dict(),
+                    "run_spec": run_spec.model_dump(),
                     "current_resource": None,
                 },
                 "force": False,
@@ -3471,7 +3472,7 @@ async def test_submits_run(
             session=session, project=project, user=user, project_role=ProjectRole.USER
         )
         submitted_at = datetime(2023, 1, 2, 3, 4, tzinfo=timezone.utc)
-        submitted_at_formatted = "2023-01-02T03:04:00+00:00"
+        submitted_at_formatted = "2023-01-02T03:04:00Z"
         last_processed_at_formatted = submitted_at_formatted
         repo = await create_repo(session=session, project_id=project.id)
         run_dict = get_dev_env_run_dict(
@@ -3517,7 +3518,7 @@ async def test_submits_run_docker_true(
             session=session, project=project, user=user, project_role=ProjectRole.USER
         )
         submitted_at = datetime(2023, 1, 2, 3, 4, tzinfo=timezone.utc)
-        submitted_at_formatted = "2023-01-02T03:04:00+00:00"
+        submitted_at_formatted = "2023-01-02T03:04:00Z"
         last_processed_at_formatted = submitted_at_formatted
         repo = await create_repo(session=session, project_id=project.id)
         run_dict = get_dev_env_run_dict(
diff --git a/src/tests/_internal/server/routers/test_users.py b/src/tests/_internal/server/routers/test_users.py
index 24af6af217..ac7ba24207 100644
--- a/src/tests/_internal/server/routers/test_users.py
+++ b/src/tests/_internal/server/routers/test_users.py
@@ -60,7 +60,7 @@ async def test_admins_see_all_non_deleted_users(
             {
                 "id": str(admin.id),
                 "username": admin.name,
-                "created_at": "2023-01-02T03:05:00+00:00",
+                "created_at": "2023-01-02T03:05:00Z",
                 "global_role": admin.global_role,
                 "email": None,
                 "active": True,
@@ -72,7 +72,7 @@ async def test_admins_see_all_non_deleted_users(
             {
                 "id": str(other_user.id),
                 "username": other_user.name,
-                "created_at": "2023-01-02T03:04:00+00:00",
+                "created_at": "2023-01-02T03:04:00Z",
                 "global_role": other_user.global_role,
                 "email": None,
                 "active": True,
@@ -117,7 +117,7 @@ async def test_returns_total_count(self, test_db, session: AsyncSession, client:
                 {
                     "id": str(admin.id),
                     "username": admin.name,
-                    "created_at": "2023-01-02T03:06:00+00:00",
+                    "created_at": "2023-01-02T03:06:00Z",
                     "global_role": admin.global_role,
                     "email": None,
                     "active": True,
@@ -160,7 +160,7 @@ async def test_paginates_results(self, test_db, session: AsyncSession, client: A
             {
                 "id": str(admin.id),
                 "username": admin.name,
-                "created_at": "2023-01-02T03:06:00+00:00",
+                "created_at": "2023-01-02T03:06:00Z",
                 "global_role": admin.global_role,
                 "email": None,
                 "active": True,
@@ -174,7 +174,7 @@ async def test_paginates_results(self, test_db, session: AsyncSession, client: A
             "/api/users/list",
             headers=get_auth_headers(admin.token),
             json={
-                "prev_created_at": "2023-01-02T03:06:00+00:00",
+                "prev_created_at": "2023-01-02T03:06:00Z",
                 "prev_id": str(admin.id),
                 "limit": 1,
             },
@@ -184,7 +184,7 @@ async def test_paginates_results(self, test_db, session: AsyncSession, client: A
             {
                 "id": str(user_one.id),
                 "username": user_one.name,
-                "created_at": "2023-01-02T03:05:00+00:00",
+                "created_at": "2023-01-02T03:05:00Z",
                 "global_role": user_one.global_role,
                 "email": None,
                 "active": True,
@@ -228,7 +228,7 @@ async def test_filters_by_name_pattern(
             {
                 "id": str(matching_user.id),
                 "username": matching_user.name,
-                "created_at": "2023-01-02T03:05:00+00:00",
+                "created_at": "2023-01-02T03:05:00Z",
                 "global_role": matching_user.global_role,
                 "email": None,
                 "active": True,
@@ -262,7 +262,7 @@ async def test_non_admins_see_only_themselves(
             {
                 "id": str(other_user.id),
                 "username": other_user.name,
-                "created_at": "2023-01-02T03:04:00+00:00",
+                "created_at": "2023-01-02T03:04:00Z",
                 "global_role": other_user.global_role,
                 "email": None,
                 "active": True,
@@ -315,7 +315,7 @@ async def test_returns_logged_in_user(
         assert response.json() == {
             "id": str(user.id),
             "username": user.name,
-            "created_at": "2023-01-02T03:04:00+00:00",
+            "created_at": "2023-01-02T03:04:00Z",
             "global_role": user.global_role,
             "email": None,
             "creds": {"token": user.token.get_plaintext_or_error()},
@@ -392,7 +392,7 @@ async def test_returns_logged_in_user(
         assert response.json() == {
             "id": str(other_user.id),
             "username": other_user.name,
-            "created_at": "2023-01-02T03:04:00+00:00",
+            "created_at": "2023-01-02T03:04:00Z",
             "global_role": other_user.global_role,
             "email": None,
             "creds": {"token": "1234"},
@@ -434,7 +434,7 @@ async def test_creates_user(self, test_db, session: AsyncSession, client: AsyncC
         assert user_data == {
             "id": "1b0e1b45-2f8c-4ab6-8010-a0d1a3e44e0e",
             "username": "test",
-            "created_at": "2023-01-02T03:04:00+00:00",
+            "created_at": "2023-01-02T03:04:00Z",
             "global_role": "user",
             "email": "test@example.com",
             "active": True,
@@ -472,7 +472,7 @@ async def test_return_400_if_username_taken(
         assert user_data == {
             "id": "1b0e1b45-2f8c-4ab6-8010-a0d1a3e44e0e",
             "username": "Test",
-            "created_at": "2023-01-02T03:04:00+00:00",
+            "created_at": "2023-01-02T03:04:00Z",
             "global_role": "user",
             "email": None,
             "active": True,
diff --git a/src/tests/_internal/server/routers/test_volumes.py b/src/tests/_internal/server/routers/test_volumes.py
index a2ea344935..e007f928c0 100644
--- a/src/tests/_internal/server/routers/test_volumes.py
+++ b/src/tests/_internal/server/routers/test_volumes.py
@@ -67,8 +67,8 @@ async def test_lists_volumes_across_projects(
                 "user": user.name,
                 "configuration": json.loads(volume2.configuration),
                 "external": False,
-                "created_at": "2023-01-02T03:05:00+00:00",
-                "last_processed_at": "2023-01-02T03:05:00+00:00",
+                "created_at": "2023-01-02T03:05:00Z",
+                "last_processed_at": "2023-01-02T03:05:00Z",
                 "status": "submitted",
                 "status_message": None,
                 "deleted": False,
@@ -86,8 +86,8 @@ async def test_lists_volumes_across_projects(
                 "user": user.name,
                 "configuration": json.loads(volume1.configuration),
                 "external": False,
-                "created_at": "2023-01-02T03:04:00+00:00",
-                "last_processed_at": "2023-01-02T03:04:00+00:00",
+                "created_at": "2023-01-02T03:04:00Z",
+                "last_processed_at": "2023-01-02T03:04:00Z",
                 "status": "submitted",
                 "status_message": None,
                 "deleted": False,
@@ -103,7 +103,7 @@ async def test_lists_volumes_across_projects(
             "/api/volumes/list",
             headers=get_auth_headers(user.token),
             json={
-                "prev_created_at": "2023-01-02T03:05:00+00:00",
+                "prev_created_at": "2023-01-02T03:05:00Z",
                 "prev_id": str(volume2.id),
             },
         )
@@ -116,8 +116,8 @@ async def test_lists_volumes_across_projects(
                 "user": user.name,
                 "configuration": json.loads(volume1.configuration),
                 "external": False,
-                "created_at": "2023-01-02T03:04:00+00:00",
-                "last_processed_at": "2023-01-02T03:04:00+00:00",
+                "created_at": "2023-01-02T03:04:00Z",
+                "last_processed_at": "2023-01-02T03:04:00Z",
                 "status": "submitted",
                 "status_message": None,
                 "deleted": False,
@@ -173,8 +173,8 @@ async def test_non_admin_cannot_see_others_projects(
                 "user": user1.name,
                 "configuration": json.loads(volume1.configuration),
                 "external": False,
-                "created_at": "2023-01-02T03:04:00+00:00",
-                "last_processed_at": "2023-01-02T03:04:00+00:00",
+                "created_at": "2023-01-02T03:04:00Z",
+                "last_processed_at": "2023-01-02T03:04:00Z",
                 "status": "submitted",
                 "status_message": None,
                 "deleted": False,
@@ -221,8 +221,8 @@ async def test_lists_volumes(self, test_db, session: AsyncSession, client: Async
                 "user": user.name,
                 "configuration": json.loads(volume.configuration),
                 "external": False,
-                "created_at": "2023-01-02T03:04:00+00:00",
-                "last_processed_at": "2023-01-02T03:04:00+00:00",
+                "created_at": "2023-01-02T03:04:00Z",
+                "last_processed_at": "2023-01-02T03:04:00Z",
                 "status": "submitted",
                 "status_message": None,
                 "deleted": False,
@@ -269,8 +269,8 @@ async def test_returns_volume(self, test_db, session: AsyncSession, client: Asyn
             "user": user.name,
             "configuration": json.loads(volume.configuration),
             "external": False,
-            "created_at": "2023-01-02T03:04:00+00:00",
-            "last_processed_at": "2023-01-02T03:04:00+00:00",
+            "created_at": "2023-01-02T03:04:00Z",
+            "last_processed_at": "2023-01-02T03:04:00Z",
             "status": "submitted",
             "status_message": None,
             "deleted": False,
@@ -321,18 +321,18 @@ async def test_creates_volume(self, test_db, session: AsyncSession, client: Asyn
             response = await client.post(
                 f"/api/project/{project.name}/volumes/create",
                 headers=get_auth_headers(user.token),
-                json={"configuration": configuration.dict()},
+                json={"configuration": configuration.model_dump()},
             )
         assert response.status_code == 200
         assert response.json() == {
             "id": "1b0e1b45-2f8c-4ab6-8010-a0d1a3e44e0e",
             "name": configuration.name,
             "project_name": project.name,
-            "configuration": configuration,
+            "configuration": configuration.model_dump(mode="json"),
             "user": user.name,
             "external": False,
-            "created_at": "2023-01-02T03:04:00+00:00",
-            "last_processed_at": "2023-01-02T03:04:00+00:00",
+            "created_at": "2023-01-02T03:04:00Z",
+            "last_processed_at": "2023-01-02T03:04:00Z",
             "status": "submitted",
             "status_message": None,
             "deleted": False,
diff --git a/src/tests/_internal/server/services/jobs/configurators/test_service.py b/src/tests/_internal/server/services/jobs/configurators/test_service.py
index a8410fcfac..7f39facac5 100644
--- a/src/tests/_internal/server/services/jobs/configurators/test_service.py
+++ b/src/tests/_internal/server/services/jobs/configurators/test_service.py
@@ -293,7 +293,7 @@ async def test_python_uses_group_python(self):
 
     async def test_user_looks_up_group_image(self, monkeypatch: pytest.MonkeyPatch):
         """When a group sets its own `image`, _user() queries that image's config."""
-        image_config = ImageConfig.parse_obj({"User": "nginx", "Entrypoint": None, "Cmd": []})
+        image_config = ImageConfig.model_validate({"User": "nginx", "Entrypoint": None, "Cmd": []})
         monkeypatch.setattr(
             "dstack._internal.server.services.jobs.configurators.base._get_image_config",
             Mock(return_value=image_config),
diff --git a/src/tests/_internal/server/services/jobs/configurators/test_task.py b/src/tests/_internal/server/services/jobs/configurators/test_task.py
index 6fab966f5e..54b8dd666d 100644
--- a/src/tests/_internal/server/services/jobs/configurators/test_task.py
+++ b/src/tests/_internal/server/services/jobs/configurators/test_task.py
@@ -47,7 +47,7 @@ async def test_adds_transport_without_credentials(self):
 
         job_spec = (await configurator.get_job_specs(replica_num=0))[0]
 
-        assert "dstack" not in job_spec.dict()
+        assert "dstack" not in job_spec.model_dump()
         assert job_spec.env == {
             "DSTACK_SERVER_URL": "http+unix://%2Frun%2Fdstack%2Fserver.sock",
         }
diff --git a/src/tests/_internal/server/services/jobs/test_jobs.py b/src/tests/_internal/server/services/jobs/test_jobs.py
index 21d69062f1..ce9b931db5 100644
--- a/src/tests/_internal/server/services/jobs/test_jobs.py
+++ b/src/tests/_internal/server/services/jobs/test_jobs.py
@@ -62,7 +62,7 @@ async def test_get_job_specs_from_run_spec_image_config_calls(
         profile=Profile(name="default"),
         ssh_key_pub="user_ssh_key",
     )
-    fake_image_config = ImageConfig.parse_obj({"Entrypoint": ["/bin/bash"]})
+    fake_image_config = ImageConfig.model_validate({"Entrypoint": ["/bin/bash"]})
     with patch(
         "dstack._internal.server.services.jobs.configurators.base._get_image_config",
         return_value=fake_image_config,
@@ -83,7 +83,7 @@ async def test_get_image_config_uses_server_default_registry(monkeypatch) -> Non
         profile=Profile(name="default"),
         ssh_key_pub="user_ssh_key",
     )
-    fake_image_config = ImageConfig.parse_obj({"Entrypoint": ["/bin/bash"]})
+    fake_image_config = ImageConfig.model_validate({"Entrypoint": ["/bin/bash"]})
     with patch(
         "dstack._internal.server.services.jobs.configurators.base._get_image_config",
         return_value=fake_image_config,
diff --git a/src/tests/_internal/server/services/requirements/test_combine.py b/src/tests/_internal/server/services/requirements/test_combine.py
index 3680161e83..d9b96da479 100644
--- a/src/tests/_internal/server/services/requirements/test_combine.py
+++ b/src/tests/_internal/server/services/requirements/test_combine.py
@@ -277,7 +277,7 @@ def test_intersection_with_duplicates(self):
         assert result == ["a", "a", "c"]
 
 
-class TestCombineIdleDuration:
+class TestCombineOptionalIdleDuration:
     def test_both_none_returns_none(self):
         assert _combine_idle_duration_optional(None, None) is None
 
diff --git a/src/tests/_internal/server/services/runner/test_client.py b/src/tests/_internal/server/services/runner/test_client.py
index e776cd14b6..efa6b64e32 100644
--- a/src/tests/_internal/server/services/runner/test_client.py
+++ b/src/tests/_internal/server/services/runner/test_client.py
@@ -84,11 +84,11 @@ def test_adds_default_project_for_server_access(self, adapter: requests_mock.Ada
         run_spec = get_run_spec(
             repo_id="repo", configuration=TaskConfiguration(commands=["true"], dstack=True)
         )
-        run = Run.construct(id=uuid.uuid4(), project_name="main", run_spec=run_spec)
-        job = Job.construct(
-            job_spec=JobSpec.construct(env={"DSTACK_TOKEN": "token"}),
+        run = Run.model_construct(id=uuid.uuid4(), project_name="main", run_spec=run_spec)
+        job = Job.model_construct(
+            job_spec=JobSpec.model_construct(env={"DSTACK_TOKEN": "token"}),
             job_submissions=[
-                JobSubmission.construct(
+                JobSubmission.model_construct(
                     id=uuid.uuid4(),
                     submitted_at=datetime.now(timezone.utc),
                 )
@@ -116,11 +116,11 @@ def test_preserves_explicit_project_for_server_access(self, adapter: requests_mo
         run_spec = get_run_spec(
             repo_id="repo", configuration=TaskConfiguration(commands=["true"], dstack=True)
         )
-        run = Run.construct(id=uuid.uuid4(), project_name="main", run_spec=run_spec)
-        job = Job.construct(
-            job_spec=JobSpec.construct(env={"DSTACK_PROJECT": "other"}),
+        run = Run.model_construct(id=uuid.uuid4(), project_name="main", run_spec=run_spec)
+        job = Job.model_construct(
+            job_spec=JobSpec.model_construct(env={"DSTACK_PROJECT": "other"}),
             job_submissions=[
-                JobSubmission.construct(
+                JobSubmission.model_construct(
                     id=uuid.uuid4(),
                     submitted_at=datetime.now(timezone.utc),
                 )
diff --git a/src/tests/_internal/server/services/runs/test_runs.py b/src/tests/_internal/server/services/runs/test_runs.py
index 0f2f906b75..cc283a6812 100644
--- a/src/tests/_internal/server/services/runs/test_runs.py
+++ b/src/tests/_internal/server/services/runs/test_runs.py
@@ -4,6 +4,7 @@
 
 from dstack._internal.core.errors import ServerClientError
 from dstack._internal.core.models.backends.base import BackendType
+from dstack._internal.core.models.duration import Duration
 from dstack._internal.core.models.profiles import Profile, ProfileRetry, RetryEvent
 from dstack._internal.core.models.runs import JobStatus, JobTerminationReason, RunStatus
 from dstack._internal.core.models.users import GlobalRole, ProjectRole
@@ -43,7 +44,7 @@ async def test_limited_list_materializes_only_latest_and_status_jobs(
             repo_id=repo.name,
             profile=Profile(
                 name="default",
-                retry=ProfileRetry(duration=3600, on_events=[RetryEvent.NO_CAPACITY]),
+                retry=ProfileRetry(duration=Duration(3600), on_events=[RetryEvent.NO_CAPACITY]),
             ),
         )
         run = await create_run(
@@ -152,7 +153,7 @@ async def test_limited_list_preserves_status_message_matrix(
         repo = await create_repo(session=session, project_id=project.id)
         retry_profile = Profile(
             name="default",
-            retry=ProfileRetry(duration=3600, on_events=[RetryEvent.NO_CAPACITY]),
+            retry=ProfileRetry(duration=Duration(3600), on_events=[RetryEvent.NO_CAPACITY]),
         )
         no_retry_profile = Profile(name="default")
 
diff --git a/src/tests/_internal/server/services/runs/test_spec.py b/src/tests/_internal/server/services/runs/test_spec.py
index 093ca768cf..0c62ad7219 100644
--- a/src/tests/_internal/server/services/runs/test_spec.py
+++ b/src/tests/_internal/server/services/runs/test_spec.py
@@ -56,7 +56,7 @@ def _service_configuration(
         data["image"] = image
     if env is not None:
         data["env"] = env
-    return ServiceConfiguration.parse_obj(data)
+    return ServiceConfiguration.model_validate(data)
 
 
 def _run_spec(configuration: ServiceConfiguration, **kwargs):
@@ -77,7 +77,7 @@ def _run_spec_with_overrides(configuration: ServiceConfiguration, **overrides) -
     )
     if not run_spec_overrides:
         return run_spec
-    return RunSpec.parse_obj({**run_spec.dict(), **run_spec_overrides})
+    return RunSpec.model_validate({**run_spec.model_dump(), **run_spec_overrides})
 
 
 class TestValidateRunSpecRetryDuration:
diff --git a/src/tests/_internal/server/services/test_docker.py b/src/tests/_internal/server/services/test_docker.py
index 45d2078dc8..14720b08aa 100644
--- a/src/tests/_internal/server/services/test_docker.py
+++ b/src/tests/_internal/server/services/test_docker.py
@@ -1,7 +1,7 @@
 import pytest
 
 import dstack._internal.server.settings as server_settings
-from dstack._internal.core.models.common import RegistryAuth
+from dstack._internal.core.models.common import RegistryAuth, validate_extra_ignore
 from dstack._internal.server.services.docker import (
     ImageConfigObject,
     ImageManifest,
@@ -98,16 +98,16 @@ def sample_image_config_object():
 
 
 def test_parse_image_manifest(sample_image_manifest):
-    ImageManifest.__response__.parse_obj(sample_image_manifest)
+    validate_extra_ignore(ImageManifest, sample_image_manifest)
 
 
 def test_parse_image_config_object(sample_image_config_object):
-    ImageConfigObject.__response__.parse_obj(sample_image_config_object)
+    validate_extra_ignore(ImageConfigObject, sample_image_config_object)
 
 
 def test_parse_image_config_object_with_config_null(sample_image_config_object):
     sample_image_config_object["config"] = None
-    config_object = ImageConfigObject.__response__.parse_obj(sample_image_config_object)
+    config_object = validate_extra_ignore(ImageConfigObject, sample_image_config_object)
     assert config_object.config is not None
 
 
@@ -121,13 +121,13 @@ def test_parse_image_config_object_with_config_null(sample_image_config_object):
 )
 def test_parse_image_config_object_user_field(sample_image_config_object, value, expected):
     sample_image_config_object["config"]["User"] = value
-    config_object = ImageConfigObject.__response__.parse_obj(sample_image_config_object)
+    config_object = validate_extra_ignore(ImageConfigObject, sample_image_config_object)
     assert config_object.config.user == expected
 
 
 def test_parse_image_config_object_user_field_missing(sample_image_config_object):
     del sample_image_config_object["config"]["User"]
-    config_object = ImageConfigObject.__response__.parse_obj(sample_image_config_object)
+    config_object = validate_extra_ignore(ImageConfigObject, sample_image_config_object)
     assert config_object.config.user is None
 
 
diff --git a/src/tests/_internal/server/services/test_fluentbit_logs.py b/src/tests/_internal/server/services/test_fluentbit_logs.py
index 937838e016..f652e0c3aa 100644
--- a/src/tests/_internal/server/services/test_fluentbit_logs.py
+++ b/src/tests/_internal/server/services/test_fluentbit_logs.py
@@ -68,8 +68,8 @@ def test_init_creates_client(self, mock_httpx_client):
     def test_write_posts_records(self, mock_httpx_client):
         writer = HTTPFluentBitWriter(host="localhost", port=8080, tag_prefix="dstack")
         records = [
-            {"message": "Hello", "@timestamp": "2023-10-06T10:00:00+00:00"},
-            {"message": "World", "@timestamp": "2023-10-06T10:00:01+00:00"},
+            {"message": "Hello", "@timestamp": "2023-10-06T10:00:00Z"},
+            {"message": "World", "@timestamp": "2023-10-06T10:00:01Z"},
         ]
         writer.write(tag="test-tag", records=records)
 
@@ -516,7 +516,7 @@ def test_read_returns_logs(self, mock_es_client):
                 "hits": [
                     {
                         "_source": {
-                            "@timestamp": "2023-10-06T10:01:53.234000+00:00",
+                            "@timestamp": "2023-10-06T10:01:53.234000Z",
                             "message": "Hello",
                             "stream": "test-stream",
                         },
@@ -524,7 +524,7 @@ def test_read_returns_logs(self, mock_es_client):
                     },
                     {
                         "_source": {
-                            "@timestamp": "2023-10-06T10:01:53.235000+00:00",
+                            "@timestamp": "2023-10-06T10:01:53.235000Z",
                             "message": "World",
                             "stream": "test-stream",
                         },
diff --git a/src/tests/_internal/server/services/test_logs.py b/src/tests/_internal/server/services/test_logs.py
index 06bfca7dea..8a8cf3394f 100644
--- a/src/tests/_internal/server/services/test_logs.py
+++ b/src/tests/_internal/server/services/test_logs.py
@@ -50,8 +50,8 @@ async def test_writes_logs(self, test_db, session: AsyncSession, tmp_path: Path)
             / "runner.log"
         )
         assert runner_log_path.read_text() == (
-            '{"timestamp":"2023-10-06T10:01:53.234000+00:00","log_source":"stdout","message":"Hello"}\n'
-            '{"timestamp":"2023-10-06T10:01:53.235000+00:00","log_source":"stdout","message":"World"}\n'
+            '{"timestamp":"2023-10-06T10:01:53.234000Z","log_source":"stdout","message":"Hello"}\n'
+            '{"timestamp":"2023-10-06T10:01:53.235000Z","log_source":"stdout","message":"World"}\n'
         )
 
     @pytest.mark.asyncio
diff --git a/src/tests/_internal/server/services/test_offers.py b/src/tests/_internal/server/services/test_offers.py
index 25ce8021ae..225853cef5 100644
--- a/src/tests/_internal/server/services/test_offers.py
+++ b/src/tests/_internal/server/services/test_offers.py
@@ -145,7 +145,7 @@ async def test_returns_az_offers(self):
             aws_offer3 = get_instance_offer_with_availability(
                 backend=BackendType.AWS, availability_zones=["az2", "az3"]
             )
-            expected_aws_offer3 = aws_offer3.copy()
+            expected_aws_offer3 = aws_offer3.model_copy()
             expected_aws_offer3.availability_zones = ["az3"]
             aws_offer4 = get_instance_offer_with_availability(
                 backend=BackendType.AWS, availability_zones=None
diff --git a/src/tests/_internal/server/services/test_repos.py b/src/tests/_internal/server/services/test_repos.py
index 50e64f84ed..a86be12427 100644
--- a/src/tests/_internal/server/services/test_repos.py
+++ b/src/tests/_internal/server/services/test_repos.py
@@ -46,7 +46,7 @@ async def _get_repo_creds(
         return None
     creds_raw = repo_creds.creds.plaintext
     assert creds_raw is not None
-    return RemoteRepoCreds.parse_raw(creds_raw)
+    return RemoteRepoCreds.model_validate_json(creds_raw)
 
 
 @pytest_asyncio.fixture
@@ -76,7 +76,7 @@ async def test_returns_none_if_repo_not_found(
             project_id=another_project.id,
             repo_name=_REPO_ID,
             repo_type=RepoType.REMOTE,
-            info=repo_info.dict(),
+            info=repo_info.model_dump(),
         )
 
         repo = await get_repo(
@@ -99,8 +99,8 @@ async def test_returns_repo_with_none_creds_if_include_creds_is_false(
             project_id=project.id,
             repo_name=_REPO_ID,
             repo_type=RepoType.REMOTE,
-            info=repo_info.dict(),
-            creds=legacy_repo_creds.dict(),
+            info=repo_info.model_dump(),
+            creds=legacy_repo_creds.model_dump(),
         )
         user_repo_creds = RemoteRepoCreds(
             clone_url="https://git.example.com/repo.git",
@@ -111,7 +111,7 @@ async def test_returns_repo_with_none_creds_if_include_creds_is_false(
             session=session,
             repo_id=repo_model.id,
             user_id=user.id,
-            creds=user_repo_creds.dict(),
+            creds=user_repo_creds.model_dump(),
         )
 
         repo = await get_repo(
@@ -134,7 +134,7 @@ async def test_returns_repo_with_none_creds_if_no_user_or_legacy_creds(
             project_id=project.id,
             repo_name=_REPO_ID,
             repo_type=RepoType.REMOTE,
-            info=repo_info.dict(),
+            info=repo_info.model_dump(),
             creds=None,
         )
         # another user's creds should be ignored
@@ -148,7 +148,7 @@ async def test_returns_repo_with_none_creds_if_no_user_or_legacy_creds(
             session=session,
             repo_id=repo_model.id,
             user_id=another_user.id,
-            creds=another_user_repo_creds.dict(),
+            creds=another_user_repo_creds.model_dump(),
         )
 
         repo = await get_repo(
@@ -193,8 +193,8 @@ async def test_returns_repo_with_user_creds_if_present(
             project_id=project.id,
             repo_name=_REPO_ID,
             repo_type=RepoType.REMOTE,
-            info=repo_info.dict(),
-            creds=legacy_repo_creds.dict() if legacy_repo_creds else None,
+            info=repo_info.model_dump(),
+            creds=legacy_repo_creds.model_dump() if legacy_repo_creds else None,
         )
         user_repo_creds = RemoteRepoCreds(
             clone_url="https://git.example.com/repo.git",
@@ -205,7 +205,7 @@ async def test_returns_repo_with_user_creds_if_present(
             session=session,
             repo_id=repo_model.id,
             user_id=user.id,
-            creds=user_repo_creds.dict(),
+            creds=user_repo_creds.model_dump(),
         )
 
         repo = await get_repo(
@@ -232,8 +232,8 @@ async def test_returns_repo_with_legacy_creds_if_user_creds_not_found(
             project_id=project.id,
             repo_name=_REPO_ID,
             repo_type=RepoType.REMOTE,
-            info=repo_info.dict(),
-            creds=legacy_repo_creds.dict(),
+            info=repo_info.model_dump(),
+            creds=legacy_repo_creds.model_dump(),
         )
 
         repo = await get_repo(
@@ -285,7 +285,7 @@ async def test_updates_repo_adding_user_creds(
             project_id=project.id,
             repo_name=_REPO_ID,
             repo_type=RepoType.REMOTE,
-            info=old_repo_info.dict(),
+            info=old_repo_info.model_dump(),
             creds=None,
         )
 
@@ -299,7 +299,7 @@ async def test_updates_repo_adding_user_creds(
         )
 
         assert repo.creds is None
-        assert RemoteRepoInfo.parse_raw(repo.info) == new_repo_info
+        assert RemoteRepoInfo.model_validate_json(repo.info) == new_repo_info
         assert await _get_repo_creds(session, repo.id, user.id) == our_repo_creds
 
     async def test_updates_repo_updating_user_creds(
@@ -311,7 +311,7 @@ async def test_updates_repo_updating_user_creds(
             project_id=project.id,
             repo_name=_REPO_ID,
             repo_type=RepoType.REMOTE,
-            info=repo_info.dict(),
+            info=repo_info.model_dump(),
             creds=None,
         )
         old_repo_creds = RemoteRepoCreds(
@@ -323,7 +323,7 @@ async def test_updates_repo_updating_user_creds(
             session=session,
             repo_id=repo.id,
             user_id=user.id,
-            creds=old_repo_creds.dict(),
+            creds=old_repo_creds.model_dump(),
         )
         new_repo_creds = RemoteRepoCreds(
             clone_url="ssh://git@git.example.com/repo.git",
@@ -356,8 +356,8 @@ async def test_updates_repo_removing_user_creds(
             project_id=project.id,
             repo_name=_REPO_ID,
             repo_type=RepoType.REMOTE,
-            info=repo_info.dict(),
-            creds=legacy_repo_creds.dict(),
+            info=repo_info.model_dump(),
+            creds=legacy_repo_creds.model_dump(),
         )
         our_repo_creds = RemoteRepoCreds(
             clone_url="https://git.example.com/repo.git",
@@ -368,7 +368,7 @@ async def test_updates_repo_removing_user_creds(
             session=session,
             repo_id=repo.id,
             user_id=user.id,
-            creds=our_repo_creds.dict(),
+            creds=our_repo_creds.model_dump(),
         )
         another_user = await _create_user(session, project, name="another-user")
         another_user_repo_creds = RemoteRepoCreds(
@@ -380,7 +380,7 @@ async def test_updates_repo_removing_user_creds(
             session=session,
             repo_id=repo.id,
             user_id=another_user.id,
-            creds=another_user_repo_creds.dict(),
+            creds=another_user_repo_creds.model_dump(),
         )
 
         repo = await init_repo(
@@ -394,7 +394,7 @@ async def test_updates_repo_removing_user_creds(
 
         # legacy creds stored in the repo are still here
         assert repo.creds is not None
-        assert RemoteRepoCreds.parse_raw(repo.creds) == legacy_repo_creds
+        assert RemoteRepoCreds.model_validate_json(repo.creds) == legacy_repo_creds
         # our personal creds are deleted
         assert await _get_repo_creds(session, repo.id, user.id) is None
         # another user's creds are still here
diff --git a/src/tests/_internal/utils/test_json_schema.py b/src/tests/_internal/utils/test_json_schema.py
deleted file mode 100644
index f6229e5354..0000000000
--- a/src/tests/_internal/utils/test_json_schema.py
+++ /dev/null
@@ -1,81 +0,0 @@
-import json
-
-from dstack._internal.core.models.configurations import DstackConfiguration, ServiceConfiguration
-from dstack._internal.core.models.profiles import ProfilesConfig
-from dstack._internal.utils.json_schema import add_extra_schema_types
-
-
-class TestAddExtraSchemaTypes:
-    def test_ref_becomes_any_of(self):
-        prop = {"$ref": "#/definitions/Foo"}
-        add_extra_schema_types(prop, extra_types=[{"type": "string"}])
-        assert prop == {"anyOf": [{"$ref": "#/definitions/Foo"}, {"type": "string"}]}
-
-    def test_all_of_keeps_first_ref_only(self):
-        prop = {"allOf": [{"$ref": "#/definitions/Foo"}]}
-        add_extra_schema_types(prop, extra_types=[{"type": "integer"}])
-        assert prop == {"anyOf": [{"$ref": "#/definitions/Foo"}, {"type": "integer"}]}
-
-    def test_any_of_is_extended_in_place(self):
-        prop = {"anyOf": [{"type": "integer"}]}
-        add_extra_schema_types(prop, extra_types=[{"type": "string"}])
-        assert prop == {"anyOf": [{"type": "integer"}, {"type": "string"}]}
-
-    def test_type_is_wrapped(self):
-        prop = {"type": "integer"}
-        add_extra_schema_types(prop, extra_types=[{"type": "string"}])
-        assert prop == {"anyOf": [{"type": "integer"}, {"type": "string"}]}
-
-    def test_other_keys_are_preserved(self):
-        prop = {"title": "Model", "description": "d", "$ref": "#/definitions/Foo"}
-        add_extra_schema_types(prop, extra_types=[{"type": "string"}])
-        assert prop["title"] == "Model"
-        assert prop["description"] == "d"
-
-    def test_discriminated_one_of_stays_grouped_with_its_discriminator(self):
-        # A `Field(discriminator=...)` union renders as `oneOf` plus a sibling `discriminator`.
-        # The two must move into the same `anyOf` member: a `discriminator` only applies to a
-        # keyword whose every member carries the tag, so flattening the extra types in beside
-        # the refs would produce an invalid schema.
-        prop = {
-            "title": "Model",
-            "oneOf": [{"$ref": "#/definitions/Foo"}, {"$ref": "#/definitions/Bar"}],
-            "discriminator": {"propertyName": "format", "mapping": {}},
-        }
-        add_extra_schema_types(prop, extra_types=[{"type": "string"}])
-        assert prop == {
-            "title": "Model",
-            "anyOf": [
-                {
-                    "oneOf": [{"$ref": "#/definitions/Foo"}, {"$ref": "#/definitions/Bar"}],
-                    "discriminator": {"propertyName": "format", "mapping": {}},
-                },
-                {"type": "string"},
-            ],
-        }
-
-    def test_one_of_without_discriminator(self):
-        prop = {"oneOf": [{"$ref": "#/definitions/Foo"}]}
-        add_extra_schema_types(prop, extra_types=[{"type": "string"}])
-        assert prop == {"anyOf": [{"oneOf": [{"$ref": "#/definitions/Foo"}]}, {"type": "string"}]}
-
-
-class TestSchemaGeneration:
-    """
-    Guards the schemas CI generates and the docs build consumes. Nothing else in the suite
-    exercises `schema_json()`, so a `schema_extra` hook that cannot handle the shape pydantic
-    emits for a field fails only in CI.
-    """
-
-    def test_dstack_configuration_schema_is_generated(self):
-        assert json.loads(DstackConfiguration.schema_json())["definitions"]
-
-    def test_profiles_config_schema_is_generated(self):
-        assert json.loads(ProfilesConfig.schema_json())["definitions"]
-
-    def test_service_model_accepts_both_the_shorthand_and_the_tagged_forms(self):
-        prop = json.loads(ServiceConfiguration.schema_json())["properties"]["model"]
-        tagged, shorthand = prop["anyOf"]
-        assert shorthand == {"type": "string"}
-        assert tagged["discriminator"]["propertyName"] == "format"
-        assert tagged["oneOf"]
diff --git a/src/tests/api/common.py b/src/tests/api/common.py
index c453b6afee..3eda9873ef 100644
--- a/src/tests/api/common.py
+++ b/src/tests/api/common.py
@@ -1,6 +1,6 @@
 import json
 from dataclasses import dataclass, field
-from typing import Any, Optional
+from typing import Any, Optional, Union
 
 import requests
 
@@ -9,13 +9,13 @@
 class RequestRecorder:
     payload: Any
     last_path: Optional[str] = None
-    last_body: Optional[str] = None
+    last_body: Optional[Union[str, bytes]] = None
     last_kwargs: dict[str, Any] = field(default_factory=dict)
 
     def __call__(
         self,
         path: str,
-        body: Optional[str] = None,
+        body: Optional[Union[str, bytes]] = None,
         raise_for_status: bool = True,
         method: str = "POST",
         **kwargs,
diff --git a/src/tests/api/test_projects.py b/src/tests/api/test_projects.py
index 38b93bae5a..6e8bb61248 100644
--- a/src/tests/api/test_projects.py
+++ b/src/tests/api/test_projects.py
@@ -3,6 +3,7 @@
 from datetime import datetime, timezone
 from uuid import UUID
 
+from dstack._internal.utils.common import render_datetime_as_api
 from dstack.api.server._projects import ProjectsAPIClient
 from tests.api.common import RequestRecorder
 
@@ -12,14 +13,14 @@
     "owner": {
         "id": "2b0e1b45-2f8c-4ab6-8010-a0d1a3e44e0e",
         "username": "u",
-        "created_at": "2023-01-02T03:04:00+00:00",
+        "created_at": "2023-01-02T03:04:00Z",
         "global_role": "user",
         "email": None,
         "active": True,
         "permissions": {"can_create_projects": True},
         "ssh_public_key": None,
     },
-    "created_at": "2023-01-02T03:04:00+00:00",
+    "created_at": "2023-01-02T03:04:00Z",
     "backends": [],
     "members": [],
     "is_public": False,
@@ -47,7 +48,7 @@ def test_projects_list_serializes_pagination_and_parses_info_list(self):
         assert payload["include_not_joined"] is True
         assert payload["return_total_count"] is True
         assert payload["name_pattern"] == "p"
-        assert payload["prev_created_at"] == dt.isoformat()
+        assert payload["prev_created_at"] == render_datetime_as_api(dt)
         assert payload["prev_id"] == str(pid)
         assert payload["limit"] == 1
         assert payload["ascending"] is True
diff --git a/src/tests/api/test_users.py b/src/tests/api/test_users.py
index c01703b811..e5c8f9afb6 100644
--- a/src/tests/api/test_users.py
+++ b/src/tests/api/test_users.py
@@ -3,13 +3,14 @@
 from datetime import datetime, timezone
 from uuid import UUID
 
+from dstack._internal.utils.common import render_datetime_as_api
 from dstack.api.server._users import UsersAPIClient
 from tests.api.common import RequestRecorder
 
 USER_PAYLOAD = {
     "id": "11111111-1111-4111-8111-111111111111",
     "username": "user",
-    "created_at": "2023-01-02T03:04:00+00:00",
+    "created_at": "2023-01-02T03:04:00Z",
     "global_role": "user",
     "email": None,
     "active": True,
@@ -38,7 +39,7 @@ def test_serializes_pagination_and_parses_info_list(self):
         assert recorder.last_path == "/api/users/list"
         assert payload["return_total_count"] is True
         assert payload["name_pattern"] == "user"
-        assert payload["prev_created_at"] == dt.isoformat()
+        assert payload["prev_created_at"] == render_datetime_as_api(dt)
         assert payload["prev_id"] == str(uid)
         assert payload["limit"] == 1
         assert payload["ascending"] is True
diff --git a/src/tests/plugins/test_rest_plugin.py b/src/tests/plugins/test_rest_plugin.py
index 7d9e35a51d..d17e58c5dd 100644
--- a/src/tests/plugins/test_rest_plugin.py
+++ b/src/tests/plugins/test_rest_plugin.py
@@ -6,7 +6,6 @@
 import pytest
 import pytest_asyncio
 import requests
-from pydantic import parse_obj_as
 from sqlalchemy.ext.asyncio import AsyncSession
 
 from dstack._internal.core.errors import ServerClientError, ServerError
@@ -44,7 +43,7 @@ async def create_run_spec(
         run_name=run_name,
         profile=profile,
         configuration=ServiceConfiguration(
-            commands=["echo hello"], port=8000, replicas=parse_obj_as(Range[int], replicas)
+            commands=["echo hello"], port=8000, replicas=Range[int].model_validate(replicas)
         ),
     )
     return spec
@@ -110,7 +109,7 @@ async def test_on_apply_plugin_service_returns_mutated_spec(
         mocker.patch.dict(os.environ, {PLUGIN_SERVICE_URI_ENV_VAR_NAME: "http://mock"})
         policy = CustomApplyPolicy()
         mock_response = Mock()
-        response_dict = {"spec": spec.dict(), "error": None}
+        response_dict = {"spec": spec.model_dump(), "error": None}
 
         if isinstance(spec, (RunSpec, FleetSpec)):
             response_dict["spec"]["profile"]["tags"] = {"env": "test", "team": "qa"}
@@ -197,7 +196,7 @@ async def test_on_apply_plugin_service_error_handling(
         mocker.patch.dict(os.environ, {PLUGIN_SERVICE_URI_ENV_VAR_NAME: "http://mock"})
         policy = CustomApplyPolicy()
         mock_response = Mock()
-        response_dict = {"spec": spec.dict(), "error": error}
+        response_dict = {"spec": spec.model_dump(), "error": error}
         mock_response.text = json.dumps(response_dict)
         mock_response.raise_for_status = Mock()
         mocker.patch("requests.post", return_value=mock_response)