Google ships Gemini 3.8 Live and Extended Thinking for real-time multimodal dialogue

Google introduced Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, calling them its most advanced live dialogue models yet. The release targets near real-time voice agents with upgrades in intelligence and parallel reasoning, real-time visual context, and background task execution that continues without interrupting the conversation. Gemini 3.8 Live is built for scale and cost efficiency, combining conversational intelligence with fluid dialogue and visual grounding. Extended Thinking is aimed at high-complexity work with increased intelligence and multi-step reasoning. Availability begins through the Gemini API, Google Workspace, and the Gemini app, with Search also listed among try surfaces.

The move extends the Gemini 3.8 family Google shipped weeks earlier with 3.8 Flash, shifting the same reasoning-first pitch into live multimodal dialogue. That matters because frontier competition is no longer only about static chat quality: hyperscalers are racing to make voice agents that can see context, keep talking, and run tools in the background. Landing the same models across consumer Gemini, Workspace, Search, and the developer API compresses the path from model release to distribution, raising the product bar for rivals that still separate assistant UX from API SKUs.

Builders should treat the pair as a latency-versus-depth tradeoff—Live for fluid, cost-sensitive conversational loops; Extended Thinking when multi-step reasoning must run while the dialogue stays open—and test interruption handling and visual grounding under real workloads. Investors tracking foundation-model platforms should read this as Google packaging agentic voice behavior as a first-class model tier, not merely an app feature.