Support for Gemini 3 model family #4911
Description
Google has just released the Gemini 3 pro model and more models will come in the weeks to come.
There is a bit of configuration required now to get it working.
Important - Thought Signatures
Gemini 3 Pro enforces stricter validation and updated handling on thought signatures which were originally introduced in Gemini 2.5 because we find that it improves performance for function calling. Thought signatures are encrypted representations of the model's internal thought process. To ensure the model maintains full context across multiple turns of a conversation, you must return the thought signatures in your subsequent requests. Improper use of thought signatures will result in a 400 error.
Thinking level: Gemini 3 Pro and later models use the thinking_level parameter to control the amount of internal reasoning the model performs (low or high) and to balance response quality, reasoning complexity, latency, and cost.
Temperature settings: If your existing code explicitly sets temperature (especially to low values for deterministic outputs), it is recommended to remove this parameter and use the Gemini 3 default of 1.0 to avoid potential looping issues or performance degradation on complex tasks.
Thought signatures: For Gemini 3 Pro and later models, if a thought signature is expected in a turn but not provided, the model returns an error instead of a warning.
Lastly, here is a list of resources: