Google launches Gemini 3.8 Live and 3.8 Live Extended Thinking: reasoning while speaking, with live switching across 97 languages

The scores are the first thing worth recording. 3.8 Live Extended Thinking captures the number one overall spot on Artificial Analysis Speech to Speech Quality Index at 82.6; it scores 68.6% on tau-Voice and 35.1% on Sierra tau-Voice-banking, which Google uses to argue for leadership in agentic task completion. On reasoning it reaches 97.7% on Big Bench Audio while maintaining what Google calls a highly competitive price point against other frontier models. 3.8 Live, meanwhile, secures second place in the Speech Agent Arena on user preference, with Google stressing its cost effectiveness at scale. On ServiceNow EVA-Bench, a benchmark for evaluating voice agents, Google says both models push the Pareto frontier for complex workflows by balancing accuracy with conversational quality, and notes specifically that this run was done on the Live API on Gemini Enterprise Agent Platform. Note the benchmark names and the platform qualifier when citing these figures - they come from Google own blog.

The second thing worth recording is engineering detail. 3.8 Live processes visual inputs in near real-time, enriching conversations with context; it automatically detects and transitions between 97 supported languages mid-conversation, so a user switching language does not need to start a new session. Easier to overlook is background execution: 3.8 Live executes tools and API calls in the background while continuing the conversation, so it can acknowledge a request and keep talking while the task finishes behind it. 3.8 Live Extended Thinking takes a different route: it reasons and speaks simultaneously. Google describes the mechanics - early verbal cues such as Let me check that acknowledge prompts naturally, and live progress narration walks users through multi-step background tasks. The intent is clear: to remove the silent gaps that have made voice agents frustrating. The blog demos include guiding employee onboarding in real time, playing chess in near real-time, turning rough sketches plus live voice feedback into functional React components, coordinating multi-step bookings with asynchronous function calls, and generating complete business plans and custom marketing toolkits on the fly by speech.

Availability and ecosystem are the third block of information. For developers, both models are available in the Gemini API and Google AI Studio; for enterprises, they enter private preview in Gemini Enterprise, coming soon to Gemini Enterprise for Customer Experience and to Google Workspace business customers. For everyone else, 3.8 Live arrives in Search Live, and 3.8 Live Extended Thinking arrives in Gemini Live, for Google AI Pro and Ultra subscribers in Workspace Docs, and for all Google AI subscribers in Gmail and Keep. Google also lists developer platforms building on the Gemini Live API: Agora, Fishjam, LangChain, LiveKit, Pipecat, Vercel and Vision Agents, which handle complex real-time media streaming infrastructure so developers can focus on the user experience. The partner quotes come from Salesforce, Genspark and Lumeris, with Google saying they highlight latency, fluidity and tool-calling capability. One caveat matters: the blog says both models are rolling out starting today, and a phased rollout means not every user sees the entry point at the same time.

Finally, two things on safety and sourcing. Google states explicitly that all audio generated by its AI products is watermarked with SynthID, an imperceptible watermark woven directly into the audio output so that AI-generated content remains detectable and can help prevent misinformation, with details of its safety and responsibility approach in the model card. On sourcing: the post is dated September 15, 2026, and the page notes it was updated September 17, 2026, so check that gap before citing benchmark figures; the page also links two related posts, on the launch of Gemini 3.8 Flash and 3.8 Flash Cyber, and on proactive cyber defence for governments and enterprises. One more caution: hourly pricing figures circulating in search results do not appear in the body of this official post, so this article deliberately omits specific price numbers and keeps only the official cost-efficiency and competitiveness wording. If you want a citing rule: take scores and availability from the official blog, and take pricing from the official pricing page.