Google announced Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking — two live dialogue AI models rolling out across Gemini API, Google AI Studio, Search Live, Gemini Live, and Google Workspace. The announcement came from Google’s Tom Ouyang and Malini Jaganathan, who described the models as “our most advanced live dialogue models yet, built for natural conversation.” Gemini 3.8 Live Extended Thinking adds the ability to reason during an active voice call, using brief verbal cues to signal reasoning rather than falling silent.
What Is Gemini 3.8 Live?
Gemini 3.8 Live is a real-time voice and multimodal dialogue model — separate from the Gemini 3.8 Flash agentic model — designed for synchronous conversation rather than async task completion. It automatically detects and switches between 97 languages mid-conversation, and it can process and discuss images and video in near real-time while speaking. Gemini 3.8 Live targets use cases such as AI-powered customer service, voice meeting assistants, and live product support.
The Extended Thinking variant adds a reasoning layer: when a caller asks a complex question — such as a policy exception or a multi-condition eligibility check — the model reasons through the answer in the background and signals it is doing so with phrases such as “Let me check that…” rather than pausing in silence. This “early verbal cue” approach is a UX design pattern for making AI reasoning feel natural in a live voice interaction.
Background Tool Execution and Workspace Integration
Both models execute tools and API calls in the background while the conversation continues, without pausing the exchange. This allows a voice AI to look up a customer record, check an order status, or call an external API during a live call — a capability that previously required a hard pause or a scripted hold.
Gemini 3.8 Live Extended Thinking is available in Google Workspace Docs for Google AI Pro and Ultra subscribers. Gemini 3.8 Live is available in Gmail and Keep for all Google AI subscribers. In Search Live, Gemini 3.8 Live is rolling out to all users; in the Gemini Live app, the Extended Thinking variant is available. Both models are accessible via the Gemini API and Google AI Studio for developers.
Note: Gemini 3.8 Live and Gemini 3.8 Flash are separate model families. Flash is optimized for agentic and asynchronous business tasks; Live is designed for real-time voice conversations. See our earlier coverage of Gemini 3.8 Flash for context on the distinction.
How Gemini 3.8 Live Compares to Competing Voice Models
Google explicitly benchmarks Gemini 3.8 Live Extended Thinking against GPT Live 1 (OpenAI), Grok Voice Think Fast 2.0 (xAI), and Astra, claiming better performance at a lower price point — though the pricing comparison has not been verified by independent sources as of this report. One source reports a benchmark score of 82.6 for Gemini 3.8 Live Extended on a voice AI benchmark; the benchmark name and its adoption status have not been independently verified and should be treated as a preliminary data point.
Our Take
Extended Thinking’s “reason while speaking” capability is the practical milestone for AI customer service: businesses can now replace hold music with an AI that signals it is working through the answer, not stalling. For multinational businesses, the automatic 97-language mid-call switching removes a structural barrier — AI customer service that works across your caller base without language-specific routing queues. The key question for enterprise buyers is how the AI Pro seat requirement maps to their existing Workspace licensing, and whether the API pricing will make sense at the call volumes of a real customer service operation.
For Context: Gemini 3.8 Live is part of a broader shift toward AI agents and voice models handling customer-facing workflows. For a guide to the current landscape of best AI agents for business — including how voice dialogue models compare to traditional task agents — see our hub coverage.
