Google has launched two live dialogue models, Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, built to make spoken interaction with artificial intelligence more fluid. Announced on 15 September 2026, both are available from launch day through the Gemini application programming interface (API), Google Workspace and the Gemini app, according to the announcement published by Google DeepMind.
The pair is aimed first at developers and enterprises, with the company describing the models as the building blocks for reliable, production-ready voice agents. Google calls them its most advanced live dialogue models to date.
Google says Gemini 3.8 Live processes visual input in near real time, automatically detects and transitions between 97 supported languages mid-conversation, and executes tools and API calls in the background while the conversation continues — so the model can acknowledge a request and keep talking while the task finishes.
Gemini 3.8 Live Extended Thinking is aimed at tasks that require deeper reasoning. According to the company, it reasons and speaks simultaneously, using early verbal cues such as "Let me check that…" to acknowledge a prompt, and narrating its progress through multi-step background tasks as they run.
Benchmark figures come from Google
Google says Gemini 3.8 Live Extended Thinking takes the top overall spot on Artificial Analysis' Speech to Speech Quality Index with a score of 82.6, and leads agentic task completion with 68.6% on τ-Voice and 35.1% on Sierra's τ-Voice-banking benchmark. The company also reports 97.7% on Big Bench Audio, and says the model maintains a competitive price point against other frontier models.
Gemini 3.8 Live took second place in the Speech Agent Arena, a result the company attributes to high user preference, and is described as cost-effective for deployment at scale. On ServiceNow's EVA-Bench, a benchmark for evaluating voice agents, Google says both models push the Pareto Frontier for complex workflows by balancing accuracy against conversational quality; the announcement notes that this test was run on the Live API on Gemini Enterprise Agent Platform.
Workspace, Search and developer platforms
In Google Workspace, the company says Gemini 3.8 Live Extended Thinking can be used in Docs Live, Gmail Live and Keep Live. Across Workspace and Search, Google positions the Live models as a more collaborative way to work through complex tasks by voice.
Developer platforms including Agora, Fishjam, LangChain, LiveKit, Pipecat, Vercel and Vision Agents build on the Gemini Live API, managing the real-time media streaming infrastructure behind voice interfaces. Google also named Salesforce, Genspark and Lumeris as partner companies, which it says have highlighted the models' latency, fluidity and tool-calling capabilities.
Demonstrations published with the announcement show Gemini 3.8 Live playing chess using visual context, guiding employee onboarding by answering live questions, and assembling business plans and marketing toolkits through speech. Extended Thinking is shown turning raw sketches and spoken feedback into functional React components, and coordinating multi-step bookings with asynchronous function calls.
All audio generated by Google's AI products is watermarked with SynthID, an imperceptible marker woven into the audio output so that AI-generated content remains detectable; the company points to the model card for details of its safety approach. Gemini 3.8 Live Extended Thinking began rolling out on the day of the announcement.