GooglemodelGemini 3.8 Live and 3.8 Live Extended Thinking
Two Google audio models that hold a spoken conversation and carry out tasks during it, one tuned for cost and one for harder multi-step work.
Details
- Both models detect and switch between 97 languages inside a single conversation.
- Extended Thinking keeps reasoning while it is still speaking, narrating progress with cues like "Let me check that", instead of going quiet to think.
- The models call tools and APIs in the background without breaking the conversation, and can respond to what a camera is showing in near real time.
- Extended Thinking scored 82.6 on Artificial Analysis' Speech to Speech Quality Index, ahead of GPT-Live-1 Astra at 81.5 and Grok Voice Think Fast 2.0 at 81.3.
- Developers get both through the Gemini API and Google AI Studio. Consumers get them in Search Live, the Gemini app, and Workspace for Pro, Ultra, and AI subscribers. Gemini Enterprise access is in private preview.
- All generated audio carries SynthID, an invisible marker that identifies it as AI-generated.
Best for
Building a voice assistant or phone-style agent that talks with someone and finishes the task in the same conversation.
When to use it
Reach for 3.8 Live on high-volume, low-latency talking, and Extended Thinking when the request needs several reasoning steps mid-call.
Who benefits
Developers and enterprise teams building customer-facing voice agents.
Join the newsletter to get ahead with AI.
I share strategies, new modules, free guides, and tools to get you better at using AI in your work.
Leading a team? See team programs →