import litellm
import asyncio

async def test_stream():
    # This call triggers the error
    response = await litellm.acompletion(
        model="scaleway/gemma-4-26b-a4b-it", 
        messages=[{"role": "user", "content": ""}],
        stream=True
    )
    async for chunk in response:
        print(chunk)

asyncio.run(test_stream())

---

When stream: false, the response is a single object. The reasoning field is present in choices[0].message.reasoning, and LiteLLM handles it correctly.


{
  "id": "chatcmpl-...",
  "object": "chat.completion",
  "choices": [{
    "index": 0,
    "message": {
      "role": "assistant",
      "content": "It looks like your message was empty...",
      "reasoning": "The user sent an empty message..." 
    },
    "finish_reason": "stop"
  }]
}


When stream: true, the reasoning field is moved into the delta object of each chunk. This specific structure causes the LiteLLM streaming iterator to return None.


data: {"id":"...","object":"chat.completion.chunk","choices":[{"index":0,"delta":{"reasoning":"Final"},"finish_reason":null}]}
data: {"id":"...","object":"chat.completion.chunk","choices":[{"index":0,"delta":{"reasoning":" check"},"finish_reason":null}]}