PR Summary

Fix #360

Update model information to support the new OpenAI models.

• The OpenAI reasoning models do not support the temperature setting -- they error out if you set the value. But the DeepSeek R1 OpenAI compatible API still allows passing the temperature setting even though it has not effect.

• I don't have access to the o3 model (requires verifying organization), but I've verified the changes with o1, o3-mini, o4-mini, gpt-4.1, gpt-4.1-mini, and gpt-4.1-nano modules and it works fine.

• The o1-mini model doesn't work because it doesn't support the system prompt (very weird). Given that the o3-mini is out and recommended, I think it's fine to not work with o1-mini.

Regarding the reasoning/thinking tokens:

• The OpenAI reasoning models do not expose the Chain of Thoughts (CoT) tokens. They may provide a summary but that is not available with the Chat Completion APIs (see details here).

• The DeepSeek R1 exposes the CoT tokens and has a really nice way of dealing with it in their version of OpenAI compatible Chat Completion APIs (see its reasoning model doc for details). However, the .NET OpenAI SDK doesn't expose reasoning_content today. The tracking issue is Additional properties from OpenAI-compatible response are needed but not exposed openai/openai-dotnet#259 (comment)

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

• The Gemini thinking models also exposes the CoT tokens but is limited with the OpenAI compatible APIs today. See https://discuss.ai.google.dev/t/reasoning-tokens-combined-with-completion-tokens-in-openai-compatibility-mode/58354/7?utm_source=chatgpt.com for details. It may support exposing CoT tokens with the same way as DeepSeek R1 in future.

We will work on supporting the CoT tokens once .NET OpenAI SDK is fixed to expose the reasoning_content field.