# Google launches Gemini 3.8 Live voice models — QANATONE > Google announced two live voice models. The Extended Thinking version can reason and run background tasks without stopping the conversation. Kaynak: https://www.qanatone.com/en/haber/gemini-3-8-live-ses-modelleri/ --- [News](https://www.qanatone.com/en/haber/) # Google launches Gemini 3.8 Live voice models 15 Sep 2026 1 min read News · Models Google announced two live voice models. The Extended Thinking version can reason and run background tasks without stopping the conversation. ## What was announced? On September 15 Google introduced two live conversation models, Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking. 3.8 Live is built for scale and cost efficiency. It pairs fluid conversation with visual grounding and can process visual input in near real time. Extended Thinking is meant for tasks that need multi-step reasoning. It can think and talk at the same time: it uses verbal cues such as "Let me check that" and describes the progress of a background task while the conversation continues. Both models detect and switch between 97 supported languages in the middle of a conversation. ## Test results According to Google's figures, Extended Thinking ranks first on Artificial Analysis' Speech to Speech Quality Index with a score of 82.6. It scored 68.6% on tau-Voice, which measures agentic task completion, 35.1% on Sierra's banking-focused tau-Voice-banking and 97.7% on Big Bench Audio. 3.8 Live placed second in Speech Agent Arena. ## Where can you use it? Both models are available to developers through the Gemini API and Google AI Studio, and they power Search Live in Google Search. They are in private preview on Gemini Enterprise. Extended Thinking is rolling out in Gemini Live in the Gemini app for Pro and Ultra subscribers and is coming to Gmail, Docs and Keep for business customers. Google did not announce pricing. All audio the models produce carries an inaudible SynthID watermark. ## What it means for businesses Whether a voice agent works depends largely on the flow of the conversation. It should not keep callers waiting, fall apart when interrupted or go silent while it checks something. Google's new models target the same problem as [GPT-Live-1](https://www.qanatone.com/en/haber/gpt-live-1-yelp-telefon-asistani/), which OpenAI opened in its API on September 10. Salesforce, ServiceNow and LiveKit are among the partners Google named. Our [latency](https://www.qanatone.com/en/nedir/latency-nedir/) explainer covers why delay matters so much in these products. ## More articles [News · Models ### Remarkable things people built with GPT-6 Astra 15 Sep 2026 · 2 min](https://www.qanatone.com/en/haber/gpt-6-astra-ile-yapilan-ilginc-seyler/)[News · Infrastructure ### Waymo plans driverless taxis in Tokyo in 2027 15 Sep 2026 · 1 min](https://www.qanatone.com/en/haber/waymo-tokyo-2027/)[News · Funding ### Anthropic heads for a second straight profitable quarter 14 Sep 2026 · 1 min](https://www.qanatone.com/en/haber/anthropic-ikinci-karli-ceyrek/)[News · Infrastructure ### Nvidia announces CUDA-Q Logical for quantum 14 Sep 2026 · 1 min](https://www.qanatone.com/en/haber/nvidia-cuda-q-logical/)[News · Policy ### Trump dismisses calls for AI guardrails 14 Sep 2026 · 1 min](https://www.qanatone.com/en/haber/trump-yapay-zeka-guvenlik-bariyerleri/)[News · Policy ### Anthropic, OpenAI and Google discuss AI testing body 14 Sep 2026 · 1 min](https://www.qanatone.com/en/haber/yapay-zeka-standart-kurulu-gorusmeleri/)[News · Policy ### Dario Amodei calls on AI labs to pace the frontier 12 Sep 2026 · 2 min](https://www.qanatone.com/en/haber/amodei-sinir-modellerde-yavaslama/)[News · Funding ### Altman says OpenAI will not go public in 2026 12 Sep 2026 · 1 min](https://www.qanatone.com/en/haber/openai-2026-halka-arz-yok/)