Google launches Gemini 3.8 Live and 3.8 Live Extended Thinking for real-time conversation

Google has released two new real-time dialogue models, Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, expanding its live audio-and-vision model line for developers, enterprises, and consumer products.

What's new

Google describes the pair as its most advanced live dialogue models to date. Gemini 3.8 Live is built for scale and cost efficiency while retaining conversational intelligence and visual grounding, making it suited to high-volume, latency-sensitive deployments. Gemini 3.8 Live Extended Thinking targets high-complexity tasks, adding increased intelligence and multi-step reasoning to the live-conversation format — the model can pause mid-conversation with verbal cues like "Let me check that…" while it works through a problem before responding.

Both models support near real-time visual processing, so they can react to what a user shows them (a screen, a document, a physical object) as the conversation unfolds. They handle 97 languages with automatic mid-conversation language switching, and can execute background tools without interrupting the flow of dialogue — for example, looking something up or taking an action while continuing to talk.

Availability is staged across three tracks: developers can access both models now through the Gemini API and Google AI Studio; enterprise customers get a private preview inside Gemini Enterprise; and general users will see the models arrive in Search Live, Gemini Live, and Google Workspace for Pro and Ultra subscribers.

Context

The release extends Google's Live model family, which the company has been iterating on quickly this year alongside its broader Gemini 3.x line. Live-format models are built specifically for low-latency, audio-first (and increasingly vision-first) conversation, distinct from the request-response pattern of standard chat models — a format Google, OpenAI, and xAI have all been racing to improve as voice and screen-sharing become standard ways users interact with AI assistants.

Why it matters

Splitting the release into a cost-efficient base model and a heavier reasoning variant mirrors a pattern now common across the industry: ship a fast, cheap default alongside an opt-in "thinking" mode for harder problems, rather than forcing every user through the same latency-cost tradeoff. For live, voice-driven interfaces specifically, the ability to reason through a multi-step task without breaking the conversational flow — rather than falling silent or handing off to a slower backend — is a meaningful usability bar that prior live models have struggled to clear. The staged rollout, from developer access now to consumer surfaces later, also signals Google is treating this as core infrastructure for Search Live and Gemini Live rather than a narrow API feature.