The Pulse
Google Makes Gemini 3.8 Live Avatars Generally Available
Google has made Gemini 3.8 Live with Live Avatar generally available in Gemini Enterprise. The service combines video avatars with speech, live visual input, background tool calls and support for 97 languages.
AI.info Team ·
Shoppers increasingly expect to describe what they need in their own words rather than work through filters and menus.
Marianne Johnson, EVP and Chief Product Officer, Cox Automotive
Gemini 3.8 Live moves into enterprise production
Google announced on September 24 that Gemini 3.8 Live with Live Avatar is generally available in Gemini Enterprise. The company says the service, first previewed at Google Cloud Next 2026, is ready for enterprise production. Customers can also build with the model through the Gemini Live API.
Live Avatar adds a visual persona to Gemini 3.8 Live’s native speech-to-speech conversation. Google says the video agents can run across web, mobile and interactive kiosks, with generated avatars whose lip movements synchronize with speech. Custom avatars are not open to all customers: Google says their creation requires allowlisting and enterprise verification.
Calls, camera feeds and 97 languages
The release bundles the avatar with capabilities for real-time interactions. Gemini 3.8 Live can process camera feeds or screen shares alongside audio, while making API and tool calls in the background and continuing to speak. Google says the model supports 97 languages and automatically detects language, including transitions during a conversation.
Google also says the speech-to-speech design can recover more naturally when users interrupt, without losing conversation context or backend transactions. The service is available through US and EU endpoints, with provisioned throughput and enterprise compliance and data-governance provisions. The announcement does not give public latency or pricing figures for the avatar feature.
Autotrader is Google’s example of the product in use
Google points to Cox Automotive’s AI shopping assistant for Autotrader as a customer application. The company says the assistant uses live screen highlighting and tool calls to help shoppers find and compare vehicles and explore financing through conversation. Johnson’s comment frames the pitch as a way to replace filters and menus with spoken requests matched to vehicle inventory.
The announcement also describes an insurance-claims demo: a caller shows damage on camera while an intake notebook fills in, and a team of agents checks the policy and prepares an adjuster packet. Google presents both examples as demonstrations of voice, video and agent actions working together, rather than as independent performance tests.
Custom avatars come with tighter controls
Customers can choose from a library of pre-built avatars, while custom avatar creation remains subject to Google’s approval process. Google says all generated audio and video streams include imperceptible SynthID watermarks. Those controls address identity and disclosure risks, but the post does not detail the verification criteria or how the watermark is surfaced to viewers.
The practical distinction is between general access to the live-avatar service and restricted access to custom likenesses. Enterprises can now test the model in Gemini Enterprise or build against its API, while organizations seeking a custom avatar still need Google’s allowlisting and verification.