September 16, 2026
Gemini live voice switches between 97 languages: Live API is now available
On September 15, 2026, Google released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, while the Live API gained support for 97 languages. The paid tier costs $0.75 per 1 million text input tokens and $0.018 per minute of audio output.
Logan Kilpatrick
@officiallogank
Say hello (literally) to Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, our new state-of-the-art live audio models with frontier-level price and performance. 3.8 Live supports 97 languages, can seamlessly switch between them, call tools asynchronously, and much more!
· 375.2K views
Previously, migrating from gemini-3.1-flash-live-preview required changing the model string and removing thinking_level or thinking_config. Now Gemini 3.8 Live Extended Thinking accepts only asynchronous tools and keeps speaking during background reasoning.
For implementation. Integration uses the Google GenAI SDK: an API key, the `gemini-3.8-live` model, and `response_modalities: ["AUDIO"]`. The Live API runs over WebSockets, while `gemini-3.8-live` executes function calls asynchronously by default.
Price and limits. Google AI Studio and the Gemini API offer free and paid tiers. An audio session lasts up to 15 minutes, and up to 2 minutes when audio and video are used simultaneously.
Both models are already available through the Google GenAI SDK with stable model IDs.