[Feature] DSPY for Reasoning Models

[TheCutestCat]

What feature would you like to see?

Hello, the most powerful current models have evolved into reasoning models, such as GPT O3, Deepseek R1, and Gemini 2 flash thinking.
Are there any research papers that validate whether DSPY methods are still effective on these types of reasoning models?"

Would you like to contribute?

• Yes, I'd like to help implement this.
• No, I just want to request it.

Additional Context

No response

[LeonSchwitzgebelKion]

Is it possible to use dspy on reasoning models such as gpt-o3? In my case, I get an error when trying to call the model. I dont set the max_tokens parameter by myself but it seems to be a standardized parameter in dspy itself. Can I put it off anyway? It blocks me from using these models.

litellm.exceptions.BadRequestError: litellm.BadRequestError: OpenAIException - Error code: 400 - {'error': {'message': "litellm.BadRequestError: AzureException BadRequestError - Unknown parameter: 'max_completion_token'. Did you mean 'max_completion_tokens'?\nReceived Model Group=o3-mini\nAvailable Model Group Fallbacks=None", 'type': None, 'param': None, 'code': '400'}}