[BUG] Use model's default temperature if not set

[qertoip]

Basic checks

• I searched existing issues - this hasn't been reported
• I can reproduce this consistently
• This is a RubyLLM bug, not my application code

What's broken?

RubyLLM sets its own temperature of 0.7 irrespective of the model.

This is not inline with the expected behavior of LLM API users.

RubyLLM default should mirror model's default.

Temperature variations radically change model behavior vs expectations.

How to reproduce

> RubyLLM.chat(model: "gemini-2.5-pro").instance_variable_get(:@temperature)
=> 0.7

Expected behavior

=> 1 because that's the Gemini's default.

What actually happened

0.7

Environment

Any

[crmne]

This seems infeasible. We would have to get that info from somewhere and add it to the model registry for every model.

Do you know of another way?

[tpaulshippy]

Maybe have a default per provider? Or at least document the 0.7 better? I agree that it is unexpected. Maybe I'm just not seeing it in the docs.

[qertoip]

This seems infeasible. We would have to get that info from somewhere and add it to the model registry for every model.

It tends to be per provider and most of them standardized on 1.0.

1.0 for OpenAI: https://platform.openai.com/docs/api-reference/responses/create

1.0 for Gemini family: https://cloud.google.com/vertex-ai/generative-ai/docs/learn/prompts/adjust-parameter-values#temperature

1.0 for Anthropic: https://docs.anthropic.com/en/api/messages#body-temperature

One should probably not pass any temperature by default to mirror actual default behavior and to avoid conflicts with top_p, top_k etc. If we want to pass temperature explicitly then 1.0 seems better default than 0.7.

[crmne]

Fixed. Models now use their own default temperatures. Thanks for catching this!