Introducing Gemini 3.8 Live and 3.8 Live Extended Thinking

Google DeepMind’s Tom Ouyang and Malini Jaganathan introduce Gemini 3.8 Live and 3.8 Live Extended Thinking as their most advanced live dialogue models yet—major upgrades in intelligence and parallel reasoning for complex voice tasks. They handle real-time visual context and background tool use without interrupting conversation, available via Gemini API, Google Workspace, and the Gemini app.

Voice assistants that think while they talk change pacing: less “please wait,” more overlapping cognition.

That’s powerful for accessibility and multitasking. It’s also easier to miss when the model acts while you’re still mid-sentence.

We’re for fluid voice AI when actions stay visible.

Live Extended Thinking’s promise—complex tasks in the background while conversation continues—is the product insight. Parallel reasoning plus visual context could make voice actually useful for cooking along with a recipe, debugging while looking at a screen, or navigating without mode switches.

Vendor claims of “most advanced yet” are expected; treat capability claims as Google-reported. The harm path: background tool calls that spend money, send messages, or change docs without a clear confirmation moment because the chat never “paused.” Extended thinking without extended consent is how agents surprise people.

I’m watching confirmation UX for side-effecting tools and latency/quality comparisons versus prior Live models in independent tests.

Announced September 15, 2026 on the DeepMind blog. Surfaces include Gemini API, Workspace, and the Gemini consumer app.