Gemini 3.8 Live is the real-time voice model in the Gemini 3.8 family, generally available on 15 September 2026 under the model IDs gemini-3.8-live and gemini-3.8-live-extended-thinking. It serves audio-to-audio conversations through the Live API with interleaved reasoning, asynchronous function calling by default and full session client content updates.
Google highlights improved voice quality, reliability and real-time agent orchestration. The Extended Thinking variant adds background reasoning during live audio interaction.