Reasoning tokens combined with completion tokens in OpenAI compatibility mode

Gemini API
,
7

The new DeepSeek R1 model has a really nice way of dealing with this in their version of OpenAI compatibility.

It would be awesome if the reasoning Gemini models could implement something similar in their OpenAI compatibility version.

reasoning_content：The content of the CoT, which is at the same level as content in the output structure.

Source: https://api-docs.deepseek.com/guides/reasoning_model

2 Likes