Reasoning detection now checks with the provider first

We changed how Agno decides whether a model supports reasoning. For Gemini, Claude, Ollama, OpenRouter, and Moonshot, it now asks the provider's API first and falls back to matching on the model id only when it has to. A model you set up for thinking can come back classified as non-reasoning if the provider says it does not support it.

Before, we inferred reasoning support from the model id alone, which sometimes got it wrong. Now the provider has the first say. We also added proper reasoning handlers for Moonshot (Kimi) and OpenRouter.

This applies on its own, with nothing for you to change.

Learn more about reasoning in the documentation.