litellm - 💡(How to fix) Fix [Bug]: Proxy model with base model support thinking but failing on completion [1 pull requests]

Check for existing issues

• I have searched the existing issues and checked that my issue is not a duplicate.

What happened?

Configuring a model in the proxy with model_info base model supported params in the proxy UI doesn't match the optional_params in the completion call.

For example, configuring a model with a name that doesn't support thinking but setting base_model as one that supports thinking works fine in the UI because the model information is coming from the base_model if provided https://github.com/BerriAI/litellm/blob/aa9e7b9808af4a46d496ca2f2436edf8ed2f5d77/litellm/proxy/proxy_server.py#L6629

For the same model, if we call completion with the thinking parameter, it fails because it is looking from the model name to get the supported parameters, and thinking is not one of them.

If the base_model is provided, it should be used to get the supported parameters, and the behavior should match the proxy model info UI

Steps to Reproduce

1. Create a proxy config with a model that doesn't support thinking (gemini-3.1-pro is not registed in the cost map today so all model info is None):

model_list:
- litellm_params:
     model: gemini/gemini-3.1-pro
   model_info:
     base_model: gemini-3.1-pro-preview

1. Start the proxy server with the config
2. Go to the models page and look for supported_openai_params for this model. thinking is in the list because base_model is being used.
3. Try to make a call to /chat/completions endpoint with "reasoning_effort": "low". Getting a validation error

{
  "error": {
    "message": "litellm.UnsupportedParamsError: gemini does not support parameters: ['reasoning_effort'], for model=gemini-3.1-pro. To drop these, set `litellm.drop_params=True` or for proxy:\n\n`litellm_settings:\n drop_params: true`\n. \n If you want to use these params dynamically send allowed_openai_params=['reasoning_effort'] in your request.. Received Model Group=gemini-3.1-pro-preview-wtp\nAvailable Model Group Fallbacks=None",
    "type": "None",
    "param": null,
    "code": "400"
  }
}

Relevant log output

What part of LiteLLM is this about?

Proxy

What LiteLLM version are you on ?

v1.83.14

Twitter / LinkedIn details

No response