[Frontend] Fix reasoning_tokens for text-based parsers in Responses API (#33513)

Signed-off-by: Jaeyeon Kim <anencore94@gmail.com>

tests/entrypoints/openai/responses/test_simple.py

@@ -134,6 +134,53 @@ async def test_streaming_output_consistency(client: OpenAI, model_name: str):
     )
 
 
+@pytest.mark.asyncio
+@pytest.mark.parametrize("model_name", [MODEL_NAME])
+async def test_streaming_reasoning_tokens_e2e(client: OpenAI, model_name: str):
+    """Verify final usage includes reasoning_tokens in streaming mode."""
+    response = await client.responses.create(
+        model=model_name,
+        input="Compute 17 * 19 and explain briefly.",
+        reasoning={"effort": "low"},
+        temperature=0.0,
+        stream=True,
+    )
+
+    completed_event = None
+    async for event in response:
+        if event.type == "response.completed":
+            completed_event = event
+
+    assert completed_event is not None
+    assert completed_event.response.status == "completed"
+    assert completed_event.response.usage is not None
+    assert completed_event.response.usage.output_tokens_details is not None
+    assert completed_event.response.usage.output_tokens_details.reasoning_tokens > 0, (
+        "Expected reasoning_tokens > 0 for streamed Qwen3 response."
+    )
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("model_name", [MODEL_NAME])
+async def test_non_streaming_reasoning_tokens_e2e(client: OpenAI, model_name: str):
+    """Verify usage includes reasoning_tokens in non-streaming mode."""
+    response = await client.responses.create(
+        model=model_name,
+        input="Compute 23 * 17 and explain briefly.",
+        reasoning={"effort": "low"},
+        temperature=0.0,
+        stream=False,
+    )
+
+    assert response is not None
+    assert response.status == "completed"
+    assert response.usage is not None
+    assert response.usage.output_tokens_details is not None
+    assert response.usage.output_tokens_details.reasoning_tokens > 0, (
+        "Expected reasoning_tokens > 0 for non-streamed Qwen3 response."
+    )
+
+
 @pytest.mark.asyncio
 @pytest.mark.parametrize("model_name", [MODEL_NAME])
 async def test_max_tokens(client: OpenAI, model_name: str):