diff --git a/packages/core/src/core/openaiContentGenerator/provider/default.test.ts b/packages/core/src/core/openaiContentGenerator/provider/default.test.ts
index 12c04c115cf..ce57b5965f2 100644
--- a/packages/core/src/core/openaiContentGenerator/provider/default.test.ts
+++ b/packages/core/src/core/openaiContentGenerator/provider/default.test.ts
@@ -425,5 +425,86 @@ describe('DefaultOpenAICompatibleProvider', () => {
       expect(result.max_tokens).toBe(8000); // GPT-4 has 16K limit, min(16K, 8K) = 8K
       expect(result).not.toHaveProperty('custom_param');
     });
+
+    it('mirrors reasoning_content into reasoning for Qwen3 assistant history turns without mutating the source request', () => {
+      const originalRequest: OpenAI.Chat.ChatCompletionCreateParams = {
+        model: 'Qwen/Qwen3.6-35B-A3B',
+        messages: [
+          { role: 'user', content: 'First turn' },
+          {
+            role: 'assistant',
+            content: 'Visible answer',
+            reasoning_content: 'Preserved chain of thought',
+          } as OpenAI.Chat.ChatCompletionAssistantMessageParam & {
+            reasoning_content: string;
+            reasoning?: string;
+          },
+          { role: 'user', content: 'Second turn' },
+        ],
+      };
+
+      const result = provider.buildRequest(originalRequest, 'prompt-id');
+      const assistant = result.messages?.[1] as {
+        reasoning_content?: string;
+        reasoning?: string;
+      };
+
+      expect(assistant.reasoning_content).toBe('Preserved chain of thought');
+      expect(assistant.reasoning).toBe('Preserved chain of thought');
+      expect(
+        (originalRequest.messages[1] as { reasoning?: string }).reasoning,
+      ).toBeUndefined();
+    });
+
+    it('does not overwrite an explicit reasoning field on Qwen3 assistant history turns', () => {
+      const originalRequest: OpenAI.Chat.ChatCompletionCreateParams = {
+        model: 'Qwen3-32B',
+        messages: [
+          {
+            role: 'assistant',
+            content: 'Visible answer',
+            reasoning_content: 'Legacy reasoning field',
+            reasoning: 'Canonical reasoning field',
+          } as OpenAI.Chat.ChatCompletionAssistantMessageParam & {
+            reasoning_content: string;
+            reasoning: string;
+          },
+        ],
+      };
+
+      const result = provider.buildRequest(originalRequest, 'prompt-id');
+      const assistant = result.messages?.[0] as {
+        reasoning_content?: string;
+        reasoning?: string;
+      };
+
+      expect(assistant.reasoning).toBe('Canonical reasoning field');
+      expect(assistant.reasoning_content).toBe('Legacy reasoning field');
+    });
+
+    it('does not mirror reasoning_content for non-Qwen3 OpenAI-compatible models', () => {
+      const originalRequest: OpenAI.Chat.ChatCompletionCreateParams = {
+        model: 'gpt-4o',
+        messages: [
+          {
+            role: 'assistant',
+            content: 'Visible answer',
+            reasoning_content: 'Preserved chain of thought',
+          } as OpenAI.Chat.ChatCompletionAssistantMessageParam & {
+            reasoning_content: string;
+            reasoning?: string;
+          },
+        ],
+      };
+
+      const result = provider.buildRequest(originalRequest, 'prompt-id');
+      const assistant = result.messages?.[0] as {
+        reasoning_content?: string;
+        reasoning?: string;
+      };
+
+      expect(assistant.reasoning_content).toBe('Preserved chain of thought');
+      expect(assistant.reasoning).toBeUndefined();
+    });
   });
 });
diff --git a/packages/core/src/core/openaiContentGenerator/provider/default.ts b/packages/core/src/core/openaiContentGenerator/provider/default.ts
index 891188dcd18..9ddd8d6b01f 100644
--- a/packages/core/src/core/openaiContentGenerator/provider/default.ts
+++ b/packages/core/src/core/openaiContentGenerator/provider/default.ts
@@ -11,6 +11,38 @@ import {
   hasExplicitOutputLimit,
 } from '../../tokenLimits.js';
 
+type AssistantMessageWithReasoningFields =
+  OpenAI.Chat.ChatCompletionAssistantMessageParam & {
+    reasoning_content?: string | null;
+    reasoning?: string | null;
+  };
+
+function shouldMirrorReasoningContentForQwen3(model: string): boolean {
+  return model.toLowerCase().includes('qwen3');
+}
+
+function mirrorReasoningContentToReasoning(
+  message: OpenAI.Chat.ChatCompletionMessageParam,
+): OpenAI.Chat.ChatCompletionMessageParam {
+  if (message.role !== 'assistant') {
+    return message;
+  }
+
+  const assistant = message as AssistantMessageWithReasoningFields;
+  if (
+    typeof assistant.reasoning_content !== 'string' ||
+    assistant.reasoning_content.length === 0 ||
+    typeof assistant.reasoning === 'string'
+  ) {
+    return message;
+  }
+
+  return {
+    ...assistant,
+    reasoning: assistant.reasoning_content,
+  } as OpenAI.Chat.ChatCompletionMessageParam;
+}
+
 /**
  * Default provider for standard OpenAI-compatible APIs
  */
@@ -75,9 +107,13 @@ export class DefaultOpenAICompatibleProvider
     // Apply output token limits to ensure max_tokens is set appropriately
     // This prevents occupying too much context window with output reservation
     const requestWithTokenLimits = this.applyOutputTokenLimit(request);
+    const messages = shouldMirrorReasoningContentForQwen3(request.model)
+      ? requestWithTokenLimits.messages.map(mirrorReasoningContentToReasoning)
+      : requestWithTokenLimits.messages;
 
     return {
       ...requestWithTokenLimits,
+      messages,
       ...(extraBody ? extraBody : {}),
     };
   }