Google launches Gemini 3.8 Live for real-time voice

Google's last Live-family voice model was Gemini 3.1 Flash Live, released in March, and the version number now matches Gemini 3.8 Flash from September 2. The main change is that 3.8 Live processes video input almost in real time and automatically detects and switches between 97 languages, including Japanese, mid-conversation. Tool execution and API calls run behind the conversation, so the model can acknowledge a request out loud and keep talking before the work finishes; Extended Thinking goes further by speaking and reasoning at once, filling gaps with phrases like "Let me check" and describing multi-step progress by voice.

The rollout spans consumer and business surfaces at once. Regular users get 3.8 Live in Search Live and Extended Thinking in Gemini Live, while Google Workspace brings Extended Thinking to Docs for Google AI Pro and Ultra subscribers and to Gmail and Keep for all Google AI plans. Developers can use the models through the Gemini API and Google AI Studio, and partners such as LiveKit, Pipecat, Agora, Vercel and LangChain can offer the Live API through their platforms. Every generated audio clip carries the SynthID watermark.

The bet here is that voice becomes a practical interface for longer tasks, not just chat. That hinges on whether the latency gains hold up in real deployments, and on how enterprises accept the undecided piece: API usage pricing is not spelled out in the official blog, so cost planning for high-volume voice agents remains an open question.