Gemini Live audio
Blog · Watch/Read · (gathered 2026-09-16)
- Simon Willison highlights Google’s release of Gemini 3.8 Live and 3.8 Live Extended Thinking, new speech-to-speech models positioned as counterparts to OpenAI’s GPT-Live family.
- He rapidly developed a browser-based UI using GPT-6 Astra Extra High to test these models, enabling interactive voice conversations with features like model interruption.
- The UI’s implementation is notably lightweight, directly connecting to Google’s
wss://generativelanguage.googleapis.comWebSocket endpoint and utilizing the Web Audio API without external libraries. - Developers can explore these new Gemini models via their direct WebSockets API, with a tutorial provided for interacting with the
BidiGenerateContentendpoint.