Google launched two voice models, Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking. Both handle spoken dialogue and started rolling out on 15 September 2026. The lighter model is aimed at scale and lower cost. It reads visual input in near real time during a conversation. It also detects and switches between 97 languages mid-conversation. Both models run tools and interface calls in the background while the conversation continues. Extended Thinking targets harder, multi-step tasks and narrates its progress aloud. Google cites first place on Artificial Analysis' speech-to-speech quality ranking for Extended Thinking. Developers get both in the Gemini API and Google AI Studio. Enterprise access is private preview in Gemini Enterprise, with customer-experience versions later.
What changed
Earlier Gemini live voice models paused the conversation while running tools or reasoning.
What it unlocks
Holding a spoken conversation while the model runs tools and narrates multi-step progress.
- 82.6 on Speech to Speech Quality Index
- 68.6% on τ-Voice
- 35.1% on τ-Voice-banking
- 97.7% on Big Bench Audio
What you need to act on it
- Gemini API or Google AI Studio access for developers
- private preview approval for Gemini Enterprise
- Google AI Pro or Ultra for Docs Live
Sources