diff --git a/packages/agent-core-v2/src/agent/llmRequester/llmRequester.ts b/packages/agent-core-v2/src/agent/llmRequester/llmRequester.ts
index 87187621ad..f394eccbb1 100644
--- a/packages/agent-core-v2/src/agent/llmRequester/llmRequester.ts
+++ b/packages/agent-core-v2/src/agent/llmRequester/llmRequester.ts
@@ -1,6 +1,7 @@
 import { createDecorator } from '#/_base/di/instantiation';
 import type { FinishReason } from '#/app/llmProtocol/finishReason';
 import type { Message, StreamedMessagePart } from '#/app/llmProtocol/message';
+import type { ThinkingEffort } from '#/app/llmProtocol/thinkingEffort';
 import type { Tool } from '#/app/llmProtocol/tool';
 import type { TokenUsage } from '#/app/llmProtocol/usage';
 import type { LLMRequestTrace } from '#/app/llmProtocol/requestTrace';
@@ -64,9 +65,15 @@ export interface LLMRequestTask {
   readonly result: Promise<LLMRequestFinish>;
 }
 
+export interface PreparedTurnRequestConfig {
+  readonly thinkingEffort: ThinkingEffort;
+}
+
 export interface IAgentLLMRequesterService {
   readonly _serviceBrand: undefined;
 
+  prepareTurnConfig(turnId: number): PreparedTurnRequestConfig | undefined;
+
   request(
     overrides?: LLMRequestOverrides,
     onPart?: LLMRequestPartHandler,
diff --git a/packages/agent-core-v2/src/agent/llmRequester/llmRequesterService.ts b/packages/agent-core-v2/src/agent/llmRequester/llmRequesterService.ts
index 6e17d74736..5ef72eb850 100644
--- a/packages/agent-core-v2/src/agent/llmRequester/llmRequesterService.ts
+++ b/packages/agent-core-v2/src/agent/llmRequester/llmRequesterService.ts
@@ -8,7 +8,10 @@
  * completion-token budget, then drives a bounded request chain: one primary
  * `model.request(input, signal)` attempt plus projection rebuilds for request
  * structure or media compatibility; general retry policy remains in the
- * loop's `stepRetry` plugin.
+ * loop's `stepRetry` plugin. When a model is configured, `prepareTurnConfig`
+ * snapshots the model, effective thinking effort, and system prompt at the turn
+ * boundary so loop telemetry and every request in that turn share one
+ * configuration.
  * Forwards streamed `part` events to the caller's `onPart`
  * handler, records `usage` through `IAgentUsageService`, resolves to an
  * `LLMRequestFinish` on the `finish` event, logs the request lifecycle
@@ -77,6 +80,7 @@ import {
   type LLMRequestSource,
   type LLMRequestTask,
   type LLMStreamTiming,
+  type PreparedTurnRequestConfig,
 } from './llmRequester';
 import type { LLMRequestTrace } from '#/app/llmProtocol/requestTrace';
 import {
@@ -152,6 +156,12 @@ export class AgentLLMRequesterService implements IAgentLLMRequesterService {
     @IEventBus private readonly eventBus: IEventBus,
   ) {}
 
+  prepareTurnConfig(turnId: number): PreparedTurnRequestConfig | undefined {
+    if (!this.profile.hasProvider()) return undefined;
+    const config = this.getOrCreateTurnConfig(turnId);
+    return { thinkingEffort: config.resolved.thinkingLevel };
+  }
+
   async request(
     overrides: LLMRequestOverrides = {},
     onPart: LLMRequestPartHandler = noopOnPart,
@@ -529,7 +539,10 @@ export class AgentLLMRequesterService implements IAgentLLMRequesterService {
 
   private resolveTurnConfig(source: LLMRequestSource | undefined): TurnRequestConfig | undefined {
     if (source?.type !== 'turn') return undefined;
-    const turnId = source.turnId;
+    return this.getOrCreateTurnConfig(source.turnId);
+  }
+
+  private getOrCreateTurnConfig(turnId: number): TurnRequestConfig {
     for (const id of this.turnConfigs.keys()) {
       if (id < turnId) this.turnConfigs.delete(id);
     }
diff --git a/packages/agent-core-v2/src/agent/loop/loopService.ts b/packages/agent-core-v2/src/agent/loop/loopService.ts
index fce65bfecb..f088bdc92c 100644
--- a/packages/agent-core-v2/src/agent/loop/loopService.ts
+++ b/packages/agent-core-v2/src/agent/loop/loopService.ts
@@ -373,9 +373,17 @@ export class AgentLoopService extends Disposable implements IAgentLoopService {
     const telemetryContext = this.telemetryContext.get();
     const turnTelemetry = this.telemetry.withContext(telemetryContext);
     const { mode, provider_type, protocol } = telemetryContext;
+    let thinkingEffort: string | undefined;
     let result: TurnResult | undefined;
     try {
-      const started: TurnStartedTelemetryEvent = { turn_id: turn.id, mode, provider_type, protocol };
+      thinkingEffort = this.llmRequester.prepareTurnConfig(turn.id)?.thinkingEffort;
+      const started: TurnStartedTelemetryEvent = {
+        turn_id: turn.id,
+        mode,
+        provider_type,
+        protocol,
+        thinking_effort: thinkingEffort,
+      };
       turnTelemetry.track2('turn_started', started);
       result = await this.run({
         turnId: turn.id,
@@ -411,6 +419,7 @@ export class AgentLoopService extends Disposable implements IAgentLoopService {
             interrupt_reason: interruptReasonFor(result),
             provider_type,
             protocol,
+            thinking_effort: thinkingEffort,
             trace_id: traceId,
           };
           turnTelemetry.track2('turn_interrupted', interrupted);
@@ -423,6 +432,7 @@ export class AgentLoopService extends Disposable implements IAgentLoopService {
         mode,
         provider_type,
         protocol,
+        thinking_effort: thinkingEffort,
         trace_id: traceId,
       };
       turnTelemetry.track2('turn_ended', ended);
diff --git a/packages/agent-core-v2/src/app/telemetry/agentTelemetryContext.ts b/packages/agent-core-v2/src/app/telemetry/agentTelemetryContext.ts
index 7ba536f4ae..8dec84014e 100644
--- a/packages/agent-core-v2/src/app/telemetry/agentTelemetryContext.ts
+++ b/packages/agent-core-v2/src/app/telemetry/agentTelemetryContext.ts
@@ -4,8 +4,8 @@
  * Agent-scoped ambient telemetry context: a per-agent property bag that domains
  * contribute to (the `plan` domain sets `mode`, the `profile` domain mirrors
  * the resolved model protocol into `provider_type` / `protocol`, the `loop`
- * domain sets `turn_id` at turn start and keeps `trace_id` at the active
- * turn's most recent request) and that turn-scoped
+ * domain sets `turn_id` at turn start and keeps `trace_id` at the active turn's
+ * most recent request) and that turn-scoped
  * telemetry snapshots at launch. Decouples turn telemetry from any
  * specific contributor so the turn domain does not need to know about plan or
  * profile. Bound at Agent scope.
diff --git a/packages/agent-core-v2/src/app/telemetry/events.ts b/packages/agent-core-v2/src/app/telemetry/events.ts
index 774fbf30b3..0323b4f451 100644
--- a/packages/agent-core-v2/src/app/telemetry/events.ts
+++ b/packages/agent-core-v2/src/app/telemetry/events.ts
@@ -47,6 +47,7 @@ export interface TurnStartedEvent {
   mode: 'agent' | 'plan';
   provider_type?: string;
   protocol?: string;
+  thinking_effort?: string;
 }
 
 export interface TurnInterruptedEvent {
@@ -56,6 +57,7 @@ export interface TurnInterruptedEvent {
   interrupt_reason: 'user_cancelled' | 'aborted' | 'max_steps' | 'error' | 'filtered' | 'blocked';
   provider_type?: string;
   protocol?: string;
+  thinking_effort?: string;
   trace_id?: string;
 }
 
@@ -66,6 +68,7 @@ export interface TurnEndedEvent {
   mode: 'agent' | 'plan';
   provider_type?: string;
   protocol?: string;
+  thinking_effort?: string;
   trace_id?: string;
 }
 
@@ -410,6 +413,7 @@ export const telemetryEventDefinitions = {
       mode: 'Agent mode the turn runs in',
       provider_type: 'Provider protocol type',
       protocol: 'Request protocol',
+      thinking_effort: 'Effective thinking effort the turn runs with',
     },
   }),
   turn_interrupted: defineTelemetryEvent<TurnInterruptedEvent>({
@@ -422,6 +426,7 @@ export const telemetryEventDefinitions = {
       interrupt_reason: 'Why the turn was interrupted',
       provider_type: 'Provider protocol type',
       protocol: 'Request protocol',
+      thinking_effort: 'Effective thinking effort the turn ran with',
       trace_id:
         'Trace id of the most recent LLM request in this turn (the failed request when the turn errored); absent for non-Kimi protocols',
     },
@@ -436,6 +441,7 @@ export const telemetryEventDefinitions = {
       mode: 'Agent mode the turn ran in',
       provider_type: 'Provider protocol type',
       protocol: 'Request protocol',
+      thinking_effort: 'Effective thinking effort the turn ran with',
       trace_id:
         'Trace id of the most recent LLM request in this turn; absent for non-Kimi protocols',
     },
diff --git a/packages/agent-core-v2/test/agent/loop/loop.test.ts b/packages/agent-core-v2/test/agent/loop/loop.test.ts
index 1f4224b6f8..6dba3eb115 100644
--- a/packages/agent-core-v2/test/agent/loop/loop.test.ts
+++ b/packages/agent-core-v2/test/agent/loop/loop.test.ts
@@ -635,7 +635,7 @@ describe('turn telemetry', () => {
 
       expect(records).toContainEqual({
         event: 'turn_started',
-        properties: { turn_id: 0, mode: 'agent', provider_type: 'kimi', protocol: 'kimi' },
+        properties: { turn_id: 0, mode: 'agent', provider_type: 'kimi', protocol: 'kimi', thinking_effort: 'off' },
       });
       expect(records).toContainEqual({
         event: 'turn_ended',
@@ -646,6 +646,7 @@ describe('turn telemetry', () => {
           mode: 'agent',
           provider_type: 'kimi',
           protocol: 'kimi',
+          thinking_effort: 'off',
         }),
       });
       expect(records.some((record) => record.event === 'turn_interrupted')).toBe(false);
@@ -654,6 +655,50 @@ describe('turn telemetry', () => {
     }
   });
 
+  it('keeps turn telemetry aligned with the request config across pre-step changes', async () => {
+    const records: TelemetryRecord[] = [];
+    const local = createTestAgent({ telemetry: recordingTelemetry(records) });
+    try {
+      const localLoop = local.get(IAgentLoopService);
+      const localProfile = local.get(IAgentProfileService);
+      local.configure({
+        modelCapabilities: {
+          image_in: false,
+          video_in: false,
+          audio_in: false,
+          thinking: true,
+          tool_use: true,
+          max_context_tokens: 1_000_000,
+        },
+      });
+      localProfile.update({ activeToolNames: [] });
+      localProfile.setThinking('on');
+      localLoop.hooks.onWillBeginStep.register('test-change-thinking', async (_ctx, next) => {
+        localProfile.setThinking('off');
+        await next();
+      });
+      local.mockNextResponse({ type: 'text', text: 'hi' });
+
+      await local.rpc.prompt({ input: [{ type: 'text', text: 'Hello' }] });
+      await local.untilTurnEnd();
+
+      const request = local.allEvents.find(
+        (event) => event.type === '[wire]' && event.event === 'llm.request',
+      );
+      expect(request?.args).toMatchObject({ thinkingEffort: 'on' });
+      expect(records).toContainEqual({
+        event: 'turn_started',
+        properties: expect.objectContaining({ turn_id: 0, thinking_effort: 'on' }),
+      });
+      expect(records).toContainEqual({
+        event: 'turn_ended',
+        properties: expect.objectContaining({ turn_id: 0, thinking_effort: 'on' }),
+      });
+    } finally {
+      await local.dispose();
+    }
+  });
+
   it('attaches the latest request trace id to turn_ended', async () => {
     const records: TelemetryRecord[] = [];
     const local = createTestAgent({ telemetry: recordingTelemetry(records) });
@@ -926,6 +971,7 @@ function createTimingRequester(): IAgentLLMRequesterService {
 
   const requester: IAgentLLMRequesterService = {
     _serviceBrand: undefined,
+    prepareTurnConfig: () => ({ thinkingEffort: 'off' }),
     async request(_overrides, onPart = () => {}) {
       await onPart({ type: 'text', text: 'answer' });
       return {
diff --git a/packages/agent-core/src/agent/turn/index.ts b/packages/agent-core/src/agent/turn/index.ts
index 3136aa6889..506f60ae0d 100644
--- a/packages/agent-core/src/agent/turn/index.ts
+++ b/packages/agent-core/src/agent/turn/index.ts
@@ -532,7 +532,7 @@ export class TurnFlow {
     const telemetryMode = this.telemetryMode();
     this.telemetryModeByTurn.set(turnId, telemetryMode);
     this.currentStepByTurn.set(turnId, 0);
-    this.agent.telemetry.track('turn_started', { turn_id: turnId, mode: telemetryMode, ...this.requestProtocolProps() });
+    this.agent.telemetry.track('turn_started', { turn_id: turnId, mode: telemetryMode, thinking_effort: this.agent.config.thinkingEffort, ...this.requestProtocolProps() });
     this.agent.fullCompaction.resetForTurn();
     this.agent.usage.beginTurn();
     this.agent.emitEvent({ type: 'turn.started', turnId, origin });
@@ -651,6 +651,7 @@ export class TurnFlow {
       reason: ended.reason,
       duration_ms: ended.durationMs,
       mode: this.telemetryModeByTurn.get(turnId) ?? this.telemetryMode(),
+      thinking_effort: this.agent.config.thinkingEffort,
       ...this.requestProtocolProps(),
       trace_id: terminalTraceId,
     });
@@ -1178,6 +1179,7 @@ export class TurnFlow {
     this.agent.telemetry.track('turn_interrupted', {
       turn_id: turnId,
       mode: this.telemetryModeByTurn.get(turnId) ?? this.telemetryMode(),
+      thinking_effort: this.agent.config.thinkingEffort,
       at_step: atStep,
       interrupt_reason: interruptReason,
       ...this.requestProtocolProps(),
diff --git a/packages/agent-core/test/agent/turn.test.ts b/packages/agent-core/test/agent/turn.test.ts
index 3af3fb3069..e3218956b2 100644
--- a/packages/agent-core/test/agent/turn.test.ts
+++ b/packages/agent-core/test/agent/turn.test.ts
@@ -519,11 +519,11 @@ describe('Agent turn flow', () => {
 
     expect(records).toContainEqual({
       event: 'turn_started',
-      properties: { turn_id: 0, mode: 'agent' },
+      properties: { turn_id: 0, mode: 'agent', thinking_effort: 'off' },
     });
     expect(records).toContainEqual({
       event: 'turn_interrupted',
-      properties: { turn_id: 0, mode: 'agent', at_step: 0, interrupt_reason: 'error' },
+      properties: { turn_id: 0, mode: 'agent', thinking_effort: 'off', at_step: 0, interrupt_reason: 'error' },
     });
   });
 
@@ -649,7 +649,7 @@ describe('Agent turn flow', () => {
     const started = records.find((candidate) => candidate.event === 'turn_started');
     expect(started).toEqual({
       event: 'turn_started',
-      properties: expect.objectContaining({ mode: 'agent', provider_type: 'kimi', protocol: 'kimi' }),
+      properties: expect.objectContaining({ mode: 'agent', provider_type: 'kimi', protocol: 'kimi', thinking_effort: 'off' }),
     });
 
     const ended = records.find((candidate) => candidate.event === 'turn_ended');
@@ -661,6 +661,7 @@ describe('Agent turn flow', () => {
         reason: 'completed',
         provider_type: 'kimi',
         protocol: 'kimi',
+        thinking_effort: 'off',
         duration_ms: expect.any(Number),
       }),
     });