Google has released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, two speech-to-speech models, according to a September 15, 2026 post by developer Simon Willison. Willison describes the models as 'a similar shape to OpenAI's GPT-Live family.' To test them, Willison used another model, GPT-6 Astra Extra High, to build a web UI from the documentation. The interface allows users to select a model and voice preset, enter an optional system prompt, and start a voice conversation through the browser, including the ability to interrupt the model while it is talking. According to Willison, the implementation uses no libraries. It connects to a WebSocket endpoint (wss://generativelanguage.googleapis.com/ws/google.ai.generativelanguage.v1alpha.GenerativeService.BidiGenerateContent) and uses the Web Audio API AudioContext for both audio capture and playback. Willison links to Google's Gemini Live tutorial for getting started with the WebSockets API. The post is written from a single independent source, and Google's own announcement details were not included in the excerpt.
- Google released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking speech-to-speech models
- The models connect via a WebSockets API and support interrupting the model mid-response
- Reported by Simon Willison on September 15, 2026, who built a no-library browser test UI
What it means for you
Google now offers voice models you can talk to in real time, where the AI listens and speaks back naturally and lets you cut in while it's talking — the same kind of conversational voice feature OpenAI already provides. This is a building block for developers, not a finished product you can just switch on.
Try this
If you build software and want to test voice interaction, follow Google's Gemini Live WebSockets tutorial and try a simple browser demo before committing to any voice feature in your product.
Who should care
Developers and technical teams building voice-driven apps, phone-assistant features, or hands-free interfaces who want a Google alternative to OpenAI's voice models.
Skip this if
You don't write code or aren't planning a real-time voice feature. There's nothing here to act on as an end user yet.
Sources: Simon Willison — read the original