diff --git a/.github/scripts/inject_uv_overrides.py b/.github/scripts/inject_uv_overrides.py
new file mode 100644
index 00000000..2e8dce30
--- /dev/null
+++ b/.github/scripts/inject_uv_overrides.py
@@ -0,0 +1,55 @@
+"""Point a downstream uv project at locally built uipath* wheels.
+
+Appends ``tool.uv.override-dependencies`` entries for every ``uipath*`` wheel
+found under ``$WHEELS_DIR`` (default: ``./wheels``) to the ``pyproject.toml``
+given as ``argv[1]``. uv overrides bypass version specifiers, so a downstream
+cap like ``uipath-langchain-client<1.19.0`` cannot mask the new code, and they
+apply to project commands (``uv sync``/``uv add``) where the ``UV_OVERRIDE``
+env var is silently ignored.
+
+An override replaces the whole requirement, dropping any extras the downstream
+graph asked for (e.g. ``uipath-langchain-client[openai]``), so EXTRAS pins the
+union of extras each overridden package must keep providing.
+"""
+
+import glob
+import os
+import pathlib
+import re
+import sys
+
+EXTRAS: dict[str, str] = {
+    "uipath-langchain-client": "all",
+}
+
+
+def main() -> None:
+    pyproject = pathlib.Path(sys.argv[1])
+    wheels = pathlib.Path(os.environ.get("WHEELS_DIR", "wheels")).resolve()
+
+    entries: list[str] = []
+    for whl in sorted(glob.glob(str(wheels / "**" / "*.whl"), recursive=True)):
+        # Wheel filename is ``{distribution}-{version}-...whl`` where the
+        # distribution escapes hyphens to underscores (uipath_llm_client ->
+        # uipath-llm-client).
+        dist = pathlib.Path(whl).name.split("-", 1)[0].replace("_", "-")
+        if not dist.startswith("uipath"):
+            continue
+        extra = f"[{EXTRAS[dist]}]" if dist in EXTRAS else ""
+        entries.append(f'    "{dist}{extra} @ {pathlib.Path(whl).as_uri()}",')
+
+    if not entries:
+        raise SystemExit(f"no uipath wheels found under {wheels}")
+
+    block = "override-dependencies = [\n" + "\n".join(entries) + "\n]\n"
+    text = pyproject.read_text()
+    if re.search(r"^\[tool\.uv\]$", text, flags=re.M):
+        text = re.sub(r"^\[tool\.uv\]\n", "[tool.uv]\n" + block, text, count=1, flags=re.M)
+    else:
+        text = text.rstrip() + "\n\n[tool.uv]\n" + block
+    pyproject.write_text(text)
+    print(f"{pyproject}:\n{block}")
+
+
+if __name__ == "__main__":
+    main()
diff --git a/.github/workflows/test-downstream-langchain.yml b/.github/workflows/test-downstream-langchain.yml
new file mode 100644
index 00000000..cd3ff29b
--- /dev/null
+++ b/.github/workflows/test-downstream-langchain.yml
@@ -0,0 +1,148 @@
+# Downstream gate: run uipath-langchain-python's tests against wheels built
+# from this PR, so a client change that breaks the downstream repo cannot merge
+# (and therefore cannot auto-publish to PyPI via cd.yml / cd-langchain.yml).
+#
+# Requires repo secrets ALPHA_TEST_CLIENT_ID / ALPHA_TEST_CLIENT_SECRET /
+# ALPHA_BASE_URL (same values as uipath-langchain-python's integration tests).
+# The `skip:downstream` label bypasses the gate for emergencies (e.g. the
+# downstream main is broken for unrelated reasons).
+name: Downstream integration
+
+on:
+  pull_request:
+    branches: [main]
+
+permissions:
+  contents: read
+
+concurrency:
+  group: downstream-${{ github.ref }}
+  cancel-in-progress: true
+
+env:
+  DOWNSTREAM_REPO: UiPath/uipath-langchain-python
+  DOWNSTREAM_REF: main
+
+jobs:
+  detect-changes:
+    runs-on: uipath-ubuntu-latest
+    outputs:
+      run_downstream: ${{ steps.detect.outputs.changed }}
+    steps:
+      - uses: actions/checkout@d23441a48e516b6c34aea4fa41551a30e30af803  # v6.1.0
+        with:
+          fetch-depth: 0
+
+      - name: Detect package source changes
+        id: detect
+        env:
+          BASE_SHA: ${{ github.event.pull_request.base.sha }}
+          HEAD_SHA: ${{ github.event.pull_request.head.sha }}
+        run: |
+          changed=$(git diff --name-only "$BASE_SHA...$HEAD_SHA" \
+            | grep -cE '^(src/|packages/uipath_langchain_client/(src/|pyproject\.toml)|pyproject\.toml|uv\.lock)' \
+            || true)
+          echo "changed=$([ "$changed" -gt 0 ] && echo true || echo false)" >> "$GITHUB_OUTPUT"
+          echo "Package source files changed: $changed"
+
+  build-wheels:
+    needs: detect-changes
+    if: needs.detect-changes.outputs.run_downstream == 'true' && !contains(github.event.pull_request.labels.*.name, 'skip:downstream')
+    runs-on: uipath-ubuntu-latest
+    steps:
+      - uses: actions/checkout@d23441a48e516b6c34aea4fa41551a30e30af803  # v6.1.0
+
+      - uses: astral-sh/setup-uv@37802adc94f370d6bfd71619e3f0bf239e1f3b78  # v7.6.0
+        with:
+          version: "0.9.27"
+
+      - name: Build wheels from the PR
+        run: |
+          uv build --out-dir wheels
+          uv build --package uipath-langchain-client --out-dir wheels
+
+      - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02  # v4.6.2
+        with:
+          name: client-wheels
+          path: wheels/*.whl
+
+  integration-tests:
+    needs: build-wheels
+    runs-on: uipath-ubuntu-latest
+    timeout-minutes: 10
+    container:
+      image: ghcr.io/astral-sh/uv:python3.12-bookworm
+      env:
+        UIPATH_JOB_KEY: "3a03d5cb-fa21-4021-894d-a8e2eda0afe0"
+        UIPATH_TRACING_ENABLED: false
+    strategy:
+      fail-fast: false
+      matrix:
+        # Only the testcases that exercise the LLM client; the downstream repo's
+        # own CI covers the SDK-scaffolding testcases and the full env matrix.
+        testcase: [chat-models, multimodal-invoke]
+        environment: [alpha]
+
+    name: "${{ matrix.testcase }} / ${{ matrix.environment }}"
+
+    steps:
+      - uses: actions/checkout@d23441a48e516b6c34aea4fa41551a30e30af803  # v6.1.0
+        with:
+          path: client
+          sparse-checkout: .github/scripts
+
+      - uses: actions/checkout@d23441a48e516b6c34aea4fa41551a30e30af803  # v6.1.0
+        with:
+          repository: ${{ env.DOWNSTREAM_REPO }}
+          ref: ${{ env.DOWNSTREAM_REF }}
+          path: downstream
+
+      - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093  # v4.3.0
+        with:
+          name: client-wheels
+          path: wheels
+
+      - name: Point testcase at the PR wheels
+        run: |
+          python3 client/.github/scripts/inject_uv_overrides.py \
+            "downstream/testcases/${{ matrix.testcase }}/pyproject.toml"
+
+      - name: Run testcase
+        working-directory: downstream/testcases/${{ matrix.testcase }}
+        env:
+          CLIENT_ID: ${{ secrets.ALPHA_TEST_CLIENT_ID }}
+          CLIENT_SECRET: ${{ secrets.ALPHA_TEST_CLIENT_SECRET }}
+          BASE_URL: ${{ secrets.ALPHA_BASE_URL }}
+        run: |
+          # With empty credentials `uipath auth` falls back to the interactive
+          # browser flow and hangs until the job timeout - fail fast instead.
+          # POSIX sh, not bash: container-job steps run under `sh -e`.
+          if [ -z "$CLIENT_ID" ] || [ -z "$CLIENT_SECRET" ] || [ -z "$BASE_URL" ]; then
+            echo "::error::ALPHA_TEST_CLIENT_ID / ALPHA_TEST_CLIENT_SECRET / ALPHA_BASE_URL secrets are not configured on this repository. Copy them from uipath-langchain-python's integration setup."
+            exit 1
+          fi
+          echo "Running testcase: ${{ matrix.testcase }} against ${{ matrix.environment }}"
+          bash run.sh
+          bash ../common/validate_output.sh
+
+  downstream-gate:
+    needs: [detect-changes, integration-tests]
+    if: always()
+    runs-on: uipath-ubuntu-latest
+    steps:
+      - name: Evaluate gate
+        run: |
+          if [[ "${{ needs.detect-changes.outputs.run_downstream }}" != "true" ]]; then
+            echo "No package source changed - downstream tests not required."
+            exit 0
+          fi
+          if [[ "${{ contains(github.event.pull_request.labels.*.name, 'skip:downstream') }}" == "true" ]]; then
+            echo "skip:downstream label present - gate bypassed."
+            exit 0
+          fi
+          if [[ "${{ needs.integration-tests.result }}" == "success" ]]; then
+            echo "Downstream green."
+            exit 0
+          fi
+          echo "Downstream tests failed - this PR would break uipath-langchain-python."
+          exit 1
diff --git a/packages/uipath_langchain_client/CHANGELOG.md b/packages/uipath_langchain_client/CHANGELOG.md
index adb94148..526b9589 100644
--- a/packages/uipath_langchain_client/CHANGELOG.md
+++ b/packages/uipath_langchain_client/CHANGELOG.md
@@ -2,6 +2,12 @@
 
 All notable changes to `uipath_langchain_client` will be documented in this file.
 
+## [1.18.4] - 2026-09-02
+
+### Fixed
+- `UiPathChatAnthropic._create`/`._acreate` now return raw responses (`with_raw_response`) to match langchain-anthropic 1.7.0, which calls `.parse()` on the result. Lifts the interim `<1.7.0` pin from 1.18.2: the `langchain-anthropic` bound on the `anthropic` and `bedrock` extras is now `>=1.7.0,<2.0.0`.
+- `UiPathChatLiteLLM.stream()`/`.astream()` no longer silently fall back to non-streaming calls on langchain-core >= 1.4, which treats the `streaming=False` that ChatLiteLLM's validator marks as explicitly set as a hard streaming opt-out.
+
 ## [1.18.3] - 2026-09-02
 
 ### Added
diff --git a/packages/uipath_langchain_client/pyproject.toml b/packages/uipath_langchain_client/pyproject.toml
index 8b3760ac..295b6ba1 100644
--- a/packages/uipath_langchain_client/pyproject.toml
+++ b/packages/uipath_langchain_client/pyproject.toml
@@ -17,12 +17,12 @@ google = [
     "langchain-google-genai>=4.2.2,<5.0.0",
 ]
 anthropic = [
-    "langchain-anthropic>=1.4.1,<1.7.0",
+    "langchain-anthropic>=1.7.0,<2.0.0",
     "anthropic[bedrock,vertex]>=0.96.0,<1.0.0",
 ]
 bedrock = [
     "langchain-aws[anthropic]>=1.4.5,<2.0.0",
-    "langchain-anthropic>=1.4.1,<1.7.0",
+    "langchain-anthropic>=1.7.0,<2.0.0",
     "anthropic[bedrock]>=0.96.0,<1.0.0",
 ]
 vertexai = [
diff --git a/packages/uipath_langchain_client/src/uipath_langchain_client/__version__.py b/packages/uipath_langchain_client/src/uipath_langchain_client/__version__.py
index f570ce1c..b91c30ed 100644
--- a/packages/uipath_langchain_client/src/uipath_langchain_client/__version__.py
+++ b/packages/uipath_langchain_client/src/uipath_langchain_client/__version__.py
@@ -1,3 +1,3 @@
 __title__ = "UiPath LangChain Client"
 __description__ = "A Python client for interacting with UiPath's LLM services via LangChain."
-__version__ = "1.18.3"
+__version__ = "1.18.4"
diff --git a/packages/uipath_langchain_client/src/uipath_langchain_client/clients/anthropic/chat_models.py b/packages/uipath_langchain_client/src/uipath_langchain_client/clients/anthropic/chat_models.py
index c1c8d92c..54dd5a38 100644
--- a/packages/uipath_langchain_client/src/uipath_langchain_client/clients/anthropic/chat_models.py
+++ b/packages/uipath_langchain_client/src/uipath_langchain_client/clients/anthropic/chat_models.py
@@ -179,14 +179,18 @@ def _async_anthropic_client(
             case _:
                 raise ValueError("Anthropic models are currently not hosted on any other provider")
 
+    # langchain-anthropic >= 1.7.0 expects _create/_acreate to return a raw-response
+    # wrapper (it calls .parse() on the result), so route through with_raw_response.
     @override
     def _create(self, payload: dict[str, Any]) -> Any:
         if "betas" in payload:
-            return self._anthropic_client.beta.messages.create(**payload)
-        return self._anthropic_client.messages.create(**payload)
+            return self._anthropic_client.beta.messages.with_raw_response.create(**payload)
+        return self._anthropic_client.messages.with_raw_response.create(**payload)
 
     @override
     async def _acreate(self, payload: dict[str, Any]) -> Any:
         if "betas" in payload:
-            return await self._async_anthropic_client.beta.messages.create(**payload)
-        return await self._async_anthropic_client.messages.create(**payload)
+            return await self._async_anthropic_client.beta.messages.with_raw_response.create(
+                **payload
+            )
+        return await self._async_anthropic_client.messages.with_raw_response.create(**payload)
diff --git a/packages/uipath_langchain_client/src/uipath_langchain_client/clients/litellm/chat_models.py b/packages/uipath_langchain_client/src/uipath_langchain_client/clients/litellm/chat_models.py
index 4bbc72b5..b779e253 100644
--- a/packages/uipath_langchain_client/src/uipath_langchain_client/clients/litellm/chat_models.py
+++ b/packages/uipath_langchain_client/src/uipath_langchain_client/clients/litellm/chat_models.py
@@ -83,6 +83,17 @@ class UiPathChatLiteLLM(UiPathBaseChatModel, ChatLiteLLM):  # type: ignore[overr
     vendor_type: VendorType | str | None = Field(default=None, exclude=True)
     api_flavor: ApiFlavor | str | None = Field(default=None, exclude=True)
 
+    def __init__(self, **kwargs: Any) -> None:
+        streaming_explicitly_set = "streaming" in kwargs
+        super().__init__(**kwargs)
+        # ChatLiteLLM's before-validator materializes a value for every field,
+        # which marks `streaming` as explicitly set. langchain-core >= 1.4
+        # treats an explicitly-set streaming=False as a hard opt-out, silently
+        # downgrading .stream()/.astream() to non-streaming invoke calls.
+        # Un-mark the field unless the caller actually passed it.
+        if not streaming_explicitly_set:
+            self.__pydantic_fields_set__.discard("streaming")
+
     # Internal core client — handles discovery, HTTPHandler lifecycle, provider resolution
     _core: UiPathLiteLLM | None = None
 
diff --git a/tests/langchain/clients/anthropic/test_raw_response_contract.py b/tests/langchain/clients/anthropic/test_raw_response_contract.py
new file mode 100644
index 00000000..efab282f
--- /dev/null
+++ b/tests/langchain/clients/anthropic/test_raw_response_contract.py
@@ -0,0 +1,290 @@
+"""Unit tests for the langchain-anthropic >= 1.7.0 raw-response contract.
+
+langchain-anthropic 1.7.0 changed `_create`/`_acreate` to return a raw-response
+wrapper and calls `.parse()` on it in `_generate`/`_agenerate`/`_stream`/`_astream`.
+These tests mock the HTTP transport and verify that the UiPath overrides (which
+route through the UiPath-constructed SDK clients) satisfy that contract for
+invoke, streaming, tool calling, structured output, and the `betas` payload
+branch — i.e. no `AttributeError: ... 'parse'` and responses parse into
+`AIMessage`s.
+"""
+
+import json
+from typing import Any
+
+import httpx
+import pytest
+from anthropic import Anthropic, AsyncAnthropic
+from langchain_core.messages import AIMessage, AIMessageChunk, HumanMessage
+from pydantic import BaseModel
+from uipath_langchain_client.clients.anthropic.chat_models import UiPathChatAnthropic
+
+from uipath.llm_client.settings import UiPathBaseSettings
+
+MODEL_NAME = "claude-sonnet-4-6"
+
+TEXT_MESSAGE: dict[str, Any] = {
+    "id": "msg_test",
+    "type": "message",
+    "role": "assistant",
+    "model": MODEL_NAME,
+    "content": [{"type": "text", "text": "Hello!"}],
+    "stop_reason": "end_turn",
+    "stop_sequence": None,
+    "usage": {"input_tokens": 10, "output_tokens": 5},
+}
+
+
+def _tool_use_message(name: str, args: dict[str, Any]) -> dict[str, Any]:
+    return {
+        **TEXT_MESSAGE,
+        "content": [{"type": "tool_use", "id": "toolu_test", "name": name, "input": args}],
+        "stop_reason": "tool_use",
+    }
+
+
+_STREAM_EVENTS: list[dict[str, Any]] = [
+    {
+        "type": "message_start",
+        "message": {
+            **TEXT_MESSAGE,
+            "content": [],
+            "stop_reason": None,
+            "usage": {"input_tokens": 10, "output_tokens": 1},
+        },
+    },
+    {"type": "content_block_start", "index": 0, "content_block": {"type": "text", "text": ""}},
+    {"type": "content_block_delta", "index": 0, "delta": {"type": "text_delta", "text": "Hel"}},
+    {"type": "content_block_delta", "index": 0, "delta": {"type": "text_delta", "text": "lo!"}},
+    {"type": "content_block_stop", "index": 0},
+    {
+        "type": "message_delta",
+        "delta": {"stop_reason": "end_turn", "stop_sequence": None},
+        "usage": {"output_tokens": 5},
+    },
+    {"type": "message_stop"},
+]
+
+SSE_BODY = "".join(
+    f"event: {event['type']}\ndata: {json.dumps(event)}\n\n" for event in _STREAM_EVENTS
+)
+
+
+class _MockedTransport:
+    """httpx.MockTransport handler that records requests and serves canned responses.
+
+    Streaming requests (``"stream": true`` in the body) get an SSE stream; other
+    requests get ``message`` as JSON. When the request declares tools, the
+    response is a ``tool_use`` block invoking the first tool with ``tool_args``.
+    """
+
+    def __init__(self, tool_args: dict[str, Any] | None = None) -> None:
+        self.requests: list[httpx.Request] = []
+        self.tool_args = tool_args or {}
+
+    def __call__(self, request: httpx.Request) -> httpx.Response:
+        self.requests.append(request)
+        body = json.loads(request.content)
+        if body.get("stream"):
+            return httpx.Response(
+                200, content=SSE_BODY, headers={"content-type": "text/event-stream"}
+            )
+        if body.get("tools"):
+            message = _tool_use_message(body["tools"][0]["name"], self.tool_args)
+        else:
+            message = TEXT_MESSAGE
+        return httpx.Response(
+            200,
+            json=message,
+            headers={"X-LangSmith-Gateway-Metadata": json.dumps({"provider": "anthropic"})},
+        )
+
+
+def _mocked_chat(
+    client_settings: UiPathBaseSettings,
+    transport_handler: _MockedTransport,
+    **model_kwargs: Any,
+) -> UiPathChatAnthropic:
+    """Build a UiPathChatAnthropic whose SDK clients run over a mock transport."""
+    params: dict[str, Any] = {
+        "model": MODEL_NAME,
+        "settings": client_settings,
+        "model_details": {},
+        "max_tokens": 1024,
+        **model_kwargs,
+    }
+    chat = UiPathChatAnthropic(**params)
+    transport = httpx.MockTransport(transport_handler)
+    # Shadow the cached properties with SDK clients over the mock transport,
+    # preserving the shape _create/_acreate rely on.
+    object.__setattr__(
+        chat,
+        "_anthropic_client",
+        Anthropic(
+            api_key="PLACEHOLDER",
+            base_url="https://mock.uipath.test",
+            max_retries=0,
+            http_client=httpx.Client(transport=transport),
+        ),
+    )
+    object.__setattr__(
+        chat,
+        "_async_anthropic_client",
+        AsyncAnthropic(
+            api_key="PLACEHOLDER",
+            base_url="https://mock.uipath.test",
+            max_retries=0,
+            http_client=httpx.AsyncClient(transport=transport),
+        ),
+    )
+    return chat
+
+
+@pytest.fixture
+def transport_handler() -> _MockedTransport:
+    return _MockedTransport(tool_args={"city": "Paris"})
+
+
+class Weather(BaseModel):
+    """Weather report for a city."""
+
+    city: str
+
+
+class TestRawResponseContract:
+    def test_invoke(
+        self, client_settings: UiPathBaseSettings, transport_handler: _MockedTransport
+    ) -> None:
+        chat = _mocked_chat(client_settings, transport_handler)
+        message = chat.invoke("Say hello")
+        assert isinstance(message, AIMessage)
+        assert message.text == "Hello!"
+        assert transport_handler.requests[-1].url.path == "/v1/messages"
+        assert "beta" not in str(transport_handler.requests[-1].url.query)
+
+    async def test_ainvoke(
+        self, client_settings: UiPathBaseSettings, transport_handler: _MockedTransport
+    ) -> None:
+        chat = _mocked_chat(client_settings, transport_handler)
+        message = await chat.ainvoke("Say hello")
+        assert isinstance(message, AIMessage)
+        assert message.text == "Hello!"
+
+    def test_stream(
+        self, client_settings: UiPathBaseSettings, transport_handler: _MockedTransport
+    ) -> None:
+        chat = _mocked_chat(client_settings, transport_handler)
+        chunks = list(chat.stream("Say hello"))
+        assert chunks
+        assert all(isinstance(chunk, AIMessageChunk) for chunk in chunks)
+        assert "".join(chunk.text for chunk in chunks) == "Hello!"
+
+    async def test_astream(
+        self, client_settings: UiPathBaseSettings, transport_handler: _MockedTransport
+    ) -> None:
+        chat = _mocked_chat(client_settings, transport_handler)
+        chunks = [chunk async for chunk in chat.astream("Say hello")]
+        assert chunks
+        assert all(isinstance(chunk, AIMessageChunk) for chunk in chunks)
+        assert "".join(chunk.text for chunk in chunks) == "Hello!"
+
+    def test_tool_calling(
+        self, client_settings: UiPathBaseSettings, transport_handler: _MockedTransport
+    ) -> None:
+        chat = _mocked_chat(client_settings, transport_handler)
+        message = chat.bind_tools([Weather]).invoke("Weather in Paris?")
+        assert isinstance(message, AIMessage)
+        assert len(message.tool_calls) == 1
+        assert message.tool_calls[0]["name"] == "Weather"
+        assert message.tool_calls[0]["args"] == {"city": "Paris"}
+
+    async def test_tool_calling_async(
+        self, client_settings: UiPathBaseSettings, transport_handler: _MockedTransport
+    ) -> None:
+        chat = _mocked_chat(client_settings, transport_handler)
+        message = await chat.bind_tools([Weather]).ainvoke("Weather in Paris?")
+        assert isinstance(message, AIMessage)
+        assert message.tool_calls[0]["name"] == "Weather"
+        assert message.tool_calls[0]["args"] == {"city": "Paris"}
+
+    def test_structured_output(
+        self, client_settings: UiPathBaseSettings, transport_handler: _MockedTransport
+    ) -> None:
+        chat = _mocked_chat(client_settings, transport_handler)
+        result = chat.with_structured_output(Weather).invoke("Weather in Paris?")
+        assert isinstance(result, Weather)
+        assert result.city == "Paris"
+
+    async def test_structured_output_async(
+        self, client_settings: UiPathBaseSettings, transport_handler: _MockedTransport
+    ) -> None:
+        chat = _mocked_chat(client_settings, transport_handler)
+        result = await chat.with_structured_output(Weather).ainvoke("Weather in Paris?")
+        assert isinstance(result, Weather)
+        assert result.city == "Paris"
+
+
+class TestBetasPayloadBranch:
+    """When the payload carries `betas`, requests must route through beta.messages."""
+
+    def test_invoke_routes_to_beta_endpoint(
+        self, client_settings: UiPathBaseSettings, transport_handler: _MockedTransport
+    ) -> None:
+        chat = _mocked_chat(
+            client_settings, transport_handler, betas=["token-efficient-tools-2025-02-19"]
+        )
+        message = chat.invoke("Say hello")
+        assert isinstance(message, AIMessage)
+        assert message.text == "Hello!"
+        request = transport_handler.requests[-1]
+        assert request.url.path == "/v1/messages"
+        assert "beta=true" in str(request.url.query)
+        assert request.headers.get("anthropic-beta") == "token-efficient-tools-2025-02-19"
+
+    async def test_ainvoke_routes_to_beta_endpoint(
+        self, client_settings: UiPathBaseSettings, transport_handler: _MockedTransport
+    ) -> None:
+        chat = _mocked_chat(
+            client_settings, transport_handler, betas=["token-efficient-tools-2025-02-19"]
+        )
+        message = await chat.ainvoke("Say hello")
+        assert isinstance(message, AIMessage)
+        request = transport_handler.requests[-1]
+        assert "beta=true" in str(request.url.query)
+        assert request.headers.get("anthropic-beta") == "token-efficient-tools-2025-02-19"
+
+    def test_stream_routes_to_beta_endpoint(
+        self, client_settings: UiPathBaseSettings, transport_handler: _MockedTransport
+    ) -> None:
+        chat = _mocked_chat(
+            client_settings, transport_handler, betas=["token-efficient-tools-2025-02-19"]
+        )
+        chunks = list(chat.stream("Say hello"))
+        assert "".join(chunk.text for chunk in chunks) == "Hello!"
+        assert "beta=true" in str(transport_handler.requests[-1].url.query)
+
+
+class TestGatewayMetadataExtraction:
+    """langchain-anthropic reads gateway metadata headers off the raw response.
+
+    `_add_gateway_metadata` only runs when the API key marks a LangSmith gateway
+    (`lsv2_` prefix); use one to verify the raw wrapper still exposes headers.
+    """
+
+    def test_gateway_metadata_lands_in_generation_info(
+        self, client_settings: UiPathBaseSettings, transport_handler: _MockedTransport
+    ) -> None:
+        chat = _mocked_chat(client_settings, transport_handler, anthropic_api_key="lsv2_test")
+        result = chat.generate([[HumanMessage("Say hello")]])
+        generation_info = result.generations[0][0].generation_info
+        assert generation_info is not None
+        assert generation_info["lc_gateway_metadata"] == {"provider": "anthropic"}
+
+    async def test_gateway_metadata_lands_in_generation_info_async(
+        self, client_settings: UiPathBaseSettings, transport_handler: _MockedTransport
+    ) -> None:
+        chat = _mocked_chat(client_settings, transport_handler, anthropic_api_key="lsv2_test")
+        result = await chat.agenerate([[HumanMessage("Say hello")]])
+        generation_info = result.generations[0][0].generation_info
+        assert generation_info is not None
+        assert generation_info["lc_gateway_metadata"] == {"provider": "anthropic"}
diff --git a/tests/langchain/clients/litellm/test_streaming_dispatch.py b/tests/langchain/clients/litellm/test_streaming_dispatch.py
new file mode 100644
index 00000000..82ca4b51
--- /dev/null
+++ b/tests/langchain/clients/litellm/test_streaming_dispatch.py
@@ -0,0 +1,50 @@
+"""Regression tests for streaming dispatch under langchain-core >= 1.4.
+
+ChatLiteLLM's before-validator materializes a value for every field, marking
+``streaming`` as explicitly set. langchain-core >= 1.4 treats an explicitly-set
+``streaming=False`` as a hard opt-out, so ``.stream()``/``.astream()`` silently
+fall back to non-streaming ``invoke`` calls. ``UiPathChatLiteLLM`` un-marks the
+field unless the caller actually passed it.
+"""
+
+from typing import Any
+
+import pytest
+from uipath_langchain_client.clients.litellm.chat_models import UiPathChatLiteLLM
+
+from uipath.llm_client.settings import UiPathBaseSettings
+
+
+@pytest.fixture
+def chat_factory(client_settings: UiPathBaseSettings, monkeypatch: pytest.MonkeyPatch) -> Any:
+    monkeypatch.setattr(
+        type(client_settings),
+        "get_model_info",
+        lambda self, *args, **kwargs: {"vendor": "openai"},
+    )
+
+    def _build(**kwargs: Any) -> UiPathChatLiteLLM:
+        return UiPathChatLiteLLM(
+            model="gpt-5.2-2025-12-11", settings=client_settings, model_details={}, **kwargs
+        )
+
+    return _build
+
+
+def test_stream_not_disabled_by_default(chat_factory: Any) -> None:
+    chat = chat_factory()
+    assert "streaming" not in chat.model_fields_set
+    assert chat._should_stream(async_api=False, stream=True)
+    assert chat._should_stream(async_api=True, stream=True)
+
+
+def test_explicit_streaming_false_still_respected(chat_factory: Any) -> None:
+    chat = chat_factory(streaming=False)
+    assert "streaming" in chat.model_fields_set
+    assert not chat._should_stream(async_api=False, stream=True)
+
+
+def test_explicit_streaming_true_kept(chat_factory: Any) -> None:
+    chat = chat_factory(streaming=True)
+    assert "streaming" in chat.model_fields_set
+    assert chat._should_stream(async_api=False)
diff --git a/uv.lock b/uv.lock
index eb5586a4..8554d0c6 100644
--- a/uv.lock
+++ b/uv.lock
@@ -166,7 +166,7 @@ wheels = [
 
 [[package]]
 name = "anthropic"
-version = "0.96.0"
+version = "0.125.0"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "anyio" },
@@ -178,9 +178,9 @@ dependencies = [
     { name = "sniffio" },
     { name = "typing-extensions" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/b9/7e/672f533dee813028d2c699bfd2a7f52c9118d7353680d9aa44b9e23f717f/anthropic-0.96.0.tar.gz", hash = "sha256:9de947b737f39452f68aa520f1c2239d44119c9b73b0fb6d4e6ca80f00279ee6", size = 658210, upload-time = "2026-04-16T14:28:02.846Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/62/f8/6f0560884b5363848347bd640b6c1d04abc25e7aa61787a232f790c6b60a/anthropic-0.125.0.tar.gz", hash = "sha256:e0cdd336580cb7411c1cdab69f80973e9bf4bff7f8e08141811d46307d45c682", size = 1112593, upload-time = "2026-08-19T22:00:42.837Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/48/5a/72f33204064b6e87601a71a6baf8d855769f8a0c1eaae8d06a1094872371/anthropic-0.96.0-py3-none-any.whl", hash = "sha256:9a6e335a354602a521cd9e777e92bfd46ba6e115bf9bbfe6135311e8fb2015b2", size = 635930, upload-time = "2026-04-16T14:28:01.436Z" },
+    { url = "https://files.pythonhosted.org/packages/2f/1a/b1bd30cda3790557e8791bec5922a6ec8fabb6fa8b008c76a39cf7be6152/anthropic-0.125.0-py3-none-any.whl", hash = "sha256:3486013602eca76d8b12540764e53654f02cf4951110bca86cf06e67428a9f21", size = 1184067, upload-time = "2026-08-19T22:00:44.596Z" },
 ]
 
 [package.optional-dependencies]
@@ -1512,16 +1512,16 @@ wheels = [
 
 [[package]]
 name = "langchain-anthropic"
-version = "1.4.1"
+version = "1.7.0"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "anthropic" },
     { name = "langchain-core" },
     { name = "pydantic" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/ca/19/55e0d3548a4d85ccee630fcfc0979af54ff4ea4f39ab0ed1ed3c6bf25f4e/langchain_anthropic-1.4.1.tar.gz", hash = "sha256:e17d027091438620e35ff2f06aefdfd63c8dcdf6abc606009ddfe3c764f2bc2e", size = 676043, upload-time = "2026-04-17T14:26:17.31Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/56/fc/52f6d1d6069bafb08626e204c89c49c8dd4a536eedbb94f0b7e78668594d/langchain_anthropic-1.7.0.tar.gz", hash = "sha256:d48e3c118ff8d3eea83f17b50234a2d2ff491a2375d565f212eb990e7e3856cb", size = 750068, upload-time = "2026-08-27T15:23:59.261Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/61/0a/20625dfea38a26e8b43654e18cafbc3595e7dc223da80f23d4fa8451dc1d/langchain_anthropic-1.4.1-py3-none-any.whl", hash = "sha256:5a48afbb2b1bad9c46badaccc8e23b0dd7ae07b7583f76bac21ddb3dac831efd", size = 49020, upload-time = "2026-04-17T14:26:15.989Z" },
+    { url = "https://files.pythonhosted.org/packages/8e/ce/e4367713080bc750e1dba845409c058f196543d1a0dc99683e2ef062f581/langchain_anthropic-1.7.0-py3-none-any.whl", hash = "sha256:68b34369aa01dad0c67bc690b8c47e09d06bd30fb28f320ad19ddc71c4445dc0", size = 60475, upload-time = "2026-08-27T15:23:57.989Z" },
 ]
 
 [[package]]
@@ -1570,10 +1570,12 @@ wheels = [
 
 [[package]]
 name = "langchain-core"
-version = "1.3.0"
+version = "1.6.1"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
+    { name = "httpx" },
     { name = "jsonpatch" },
+    { name = "langchain-protocol" },
     { name = "langsmith" },
     { name = "packaging" },
     { name = "pydantic" },
@@ -1582,9 +1584,9 @@ dependencies = [
     { name = "typing-extensions" },
     { name = "uuid-utils" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/92/fe/20190232d9b513242899dbb0c2bb77e31b4d61e343743adbe90ebc2603d2/langchain_core-1.3.0.tar.gz", hash = "sha256:14a39f528bf459aa3aa40d0a7f7f1bae7520d435ef991ae14a4ceb74d8c49046", size = 860755, upload-time = "2026-04-17T14:51:38.298Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/90/12/aff76ca89c219ebe6f9dd3c5dbc4e3b1cf5450e9fc7037dccad23d45cd7a/langchain_core-1.6.1.tar.gz", hash = "sha256:1b156cb395aac4f009a8a1b38a574c7d948fe2d5f74c96e0d8a5017b4149e04f", size = 1003359, upload-time = "2026-08-27T19:31:14.956Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/f8/e2/dbfa347aa072a6dc4cd38d6f9ebfc730b4c14c258c47f480f4c5c546f177/langchain_core-1.3.0-py3-none-any.whl", hash = "sha256:baf16ee028475df177b9ab8869a751c79406d64a6f12125b93802991b566cced", size = 515140, upload-time = "2026-04-17T14:51:36.274Z" },
+    { url = "https://files.pythonhosted.org/packages/8e/25/f50dd65673c819aa33d3c34df58c115dbb6ec627d19f93e6e401dd0fc8d7/langchain_core-1.6.1-py3-none-any.whl", hash = "sha256:954a84132a5cb0435d27b910e336347b6744ecc18fbeef1e2de7029a0959841a", size = 571478, upload-time = "2026-08-27T19:31:13.34Z" },
 ]
 
 [[package]]
@@ -1670,6 +1672,18 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/2c/7b/e8c3beeab0ca042529533072ebee69c66327c1805b3133531b58c422baab/langchain_openai-1.2.0-py3-none-any.whl", hash = "sha256:b3ed14dc48e40890605136f26c6b07e8f293987d95e734ab67cbfa572c523456", size = 98592, upload-time = "2026-04-23T00:43:34.135Z" },
 ]
 
+[[package]]
+name = "langchain-protocol"
+version = "0.0.19"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "typing-extensions" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/14/56/913599f2f9cec8524868929f12d72b2ede377a6056ca8a40a32bdadfa535/langchain_protocol-0.0.19.tar.gz", hash = "sha256:79d90a1425122ac87e8052e2ec054fbd09c3edbf341bdfb6397112a495c7bf8c", size = 6265, upload-time = "2026-08-26T21:12:00.703Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/80/c9/f6cbf357d48ccbd18bb394433b1fd7ad9be004eed9377ad08bb85777e5e6/langchain_protocol-0.0.19-py3-none-any.whl", hash = "sha256:4cdf879a492a35980fd859ae792d3c65458ccaae504e183c9a10d7eac1f0720f", size = 7327, upload-time = "2026-08-26T21:11:59.781Z" },
+]
+
 [[package]]
 name = "langchain-tests"
 version = "1.1.6"
@@ -3774,9 +3788,9 @@ requires-dist = [
     { name = "anthropic", extras = ["bedrock", "vertex"], marker = "extra == 'all'", specifier = ">=0.96.0,<1.0.0" },
     { name = "anthropic", extras = ["bedrock", "vertex"], marker = "extra == 'anthropic'", specifier = ">=0.96.0,<1.0.0" },
     { name = "langchain", specifier = ">=1.2.15,<2.0.0" },
-    { name = "langchain-anthropic", marker = "extra == 'all'", specifier = ">=1.4.1,<1.7.0" },
-    { name = "langchain-anthropic", marker = "extra == 'anthropic'", specifier = ">=1.4.1,<1.7.0" },
-    { name = "langchain-anthropic", marker = "extra == 'bedrock'", specifier = ">=1.4.1,<1.7.0" },
+    { name = "langchain-anthropic", marker = "extra == 'all'", specifier = ">=1.7.0,<2.0.0" },
+    { name = "langchain-anthropic", marker = "extra == 'anthropic'", specifier = ">=1.7.0,<2.0.0" },
+    { name = "langchain-anthropic", marker = "extra == 'bedrock'", specifier = ">=1.7.0,<2.0.0" },
     { name = "langchain-aws", extras = ["anthropic"], marker = "extra == 'all'", specifier = ">=1.4.5,<2.0.0" },
     { name = "langchain-aws", extras = ["anthropic"], marker = "extra == 'bedrock'", specifier = ">=1.4.5,<2.0.0" },
     { name = "langchain-azure-ai", marker = "extra == 'all'", specifier = ">=1.2.2,<2.0.0" },