Gemini 3.8 Live can reason and talk to you at the same time

What happened: Google released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking on 15 September 2026, two native speech-to-speech models built for real-time voice agents.

The details: Standard 3.8 Live targets speed and cost at scale. Extended Thinking adds multi-step reasoning while speaking, with low, medium and high reasoning settings. Rather than going silent to think, the model acknowledges out loud, keeps talking, runs tools in the background, and reports results when done.

The numbers: Extended Thinking (High) scored 82.6 on the Artificial Analysis Speech to Speech Index, ahead of OpenAI's GPT-Live-1 (Astra, medium) at 81.5 and xAI's Grok Voice Think Fast 2.0 High at 81.3. It hit 68.6% on Tau-Voice versus 30.1% for standard 3.8 Live, and 97.7% on Big Bench Audio. Time to first audio is 1.18 seconds for standard Live and 1.35 seconds for Extended Thinking High, down from 2.99 seconds on Gemini 3.1 Flash Live.

Who's affected: Developers building customer service bots and voice assistants, via the Gemini API and Google AI Studio. Extended Thinking is also coming to Gemini Live, Docs, Gmail and Keep for subscribers.

The catch: Both models are hosted only. Audio input costs $0.005 a minute, output $0.018 a minute. All generated audio carries a SynthID watermark.

The rivals: Google's launch puts it ahead of OpenAI and xAI on the speech-to-speech index, at least for now.

The context: The models switch between 97 languages mid-call and are supported by Agora, LangChain, LiveKit, Pipecat and Vercel.

Why it matters: Voice agents no longer need to pause and go quiet to think. They can do real work, like checking a database or booking something, while staying in conversation.

What's next: Extended Thinking's rollout into Gmail, Docs and Keep will extend reasoning-while-talking beyond standalone voice agents into everyday productivity tools.