[Feature] reasoning_content in API for reasoning models

[yasirerkam]

To better support reasoning models, adding a reasoning_content parameter to the API would be great so that users can see the steps in the reasoning process.

tool-calls

from openai import OpenAI
client = OpenAI(api_key="", base_url="https://api.deepseek.com")

messages = [{"role": "user", "content": "9.11 and 9.8, which is greater?"}]
response = client.chat.completions.create(
model="deepseek-reasoner",
messages=messages
)

+ reasoning_content = response.choices[0].message.reasoning_content
content = response.choices[0].message.content

[stickperson]

FYI this is preventing the deepseek-reasoner model from working.

Ran into this error: Request failed: Bad request (400): Missing reasoning_content field in the assistant message at message index 2. For more information, please refer to https://api-docs.deepseek.com/guides/thinking_mode#tool-calls.

[stickperson]

@Abhijay007 I started digging into this a bit. Happy to give it a go if you do not have time.

[Abhijay007]

Sure @stickperson go for it :)

[stickperson]

@Abhijay007 ran into a bit of a wall and may not have as much time as originally though to dig deeper. I pushed my changes to https://github.com/stickperson/goose/tree/openai-reasoning-content if you want a starting point.

Some message with the assistant role does not have reasoning_content still. It may have something to do with how I am processing responses.

[Abhijay007]

Thanks for this, @stickperson. I will take it from here. Thanks for all the effort you put into it :)

[yasirerkam]

Perhaps it would be helpful to look at how it was handled in the other project:
link1
link2