Types for ResponseReasoningPartAdded, ResponseReasoningPartDone

[qandrew]

Confirm this is a feature request for the Python library and not the underlying OpenAI API.

• This is a feature request for the Python library

Describe the feature or improvement you're requesting

Hi, I am working on supporting streaming in GPT-OSS in vLLM. With GPT-OSS, we can show the full reasoning text (as opposed to the summary version). However, I think there are some gaps in the OpenAI library for supporting streaming.

With GPT-5 API, the stream response would look like

ReasoningSummaryPartAdded
ReasoningSummaryTextDelta
ReasoningSummaryTextDelta
…
ReasoningSummaryTextDone
ReasoningSummaryPartDone

But in GPTOSS, we could get ResponseReasoningText. So it should look like

ResponseReasoningPartAdded  # or should it be ResponseContentPartAdded?
ResponseReasoningTextDelta
ResponseReasoningTextDelta
…
ResponseReasoningTextDone
ResponseReasoningPartDone

The types ResponseReasoningPartAdded, ResponseReasoningPartDone don't currently exist.

ResponseContentPartDone currently does not support Part as a ResponseReasoningItem (https://github.com/openai/openai-python/blob/main/src/openai/types/responses/response_content_part_done_event.py#L13).

So I'm wondering, would the path forward be to support ResponseReasoningPartAdded, ResponseReasoningPartDone? (or amend ResponseContentPartDone?)

Thanks in advance!

Additional context

vllm-project/vllm#24938
https://community.openai.com/t/purpose-of-response-reasoning-text/1344924/3

[qandrew]

I'd be happy to implement this, as it seems pretty straightforward

[nomiveritas]

Suggested solution for Python library streaming:

Introduce two new dataclasses: ResponseReasoningPartAdded and ResponseReasoningPartDone to store reasoning text parts during streaming.

Alternatively, extend ResponseContentPartDone to support per-part reasoning elements if creating new types is unnecessary.

Ensure streaming parts are wrapped in these classes and concatenated on the client side.

Test that all parts arrive correctly with no missing or duplicate sections, and that the full text matches GPT-OSS output.