Gemini 3.8 Live Extended Thinking
This live model adds background reasoning to a continuous audio interaction.
What the release provides
Background thinking supports low, medium and high levels with summaries. Function calls are asynchronous only. The client can monitor interaction_status while the model continues to handle audio.Gemini 3.8 Live Extended Thinking official API documentation
How to use it well
Practical guidance. Use it for conversations where a speaker can continue while a difficult question is processed. Keep pending actions visible in application state. If a user corrects the request mid-action, ensure the tool layer applies that change rather than executing stale arguments.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | API |
| Model identifiers | gemini-3.8-live-extended-thinking |
| Context window | 131,072 tokens API input limit; reserve output capacity where the service shares one window. Gemini 3.8 Live Extended Thinking official API documentation |
| Maximum output | 65,536 tokens Published API output limit; request settings can be lower. Gemini 3.8 Live Extended Thinking official API documentation |