Google releases Gemini 3.8 Live with Live Avatar for enterprise customers

Google is rolling out Gemini 3.8 Live with live video avatar capability inside Gemini Enterprise, one week after announcing the model itself and its Extended Thinking variant. Fabien Blanc-paques, group product manager for Gemini Live, said the technology — first demonstrated at Google Cloud Next 2026 — is now ready for enterprise production with endpoints in the United States and the European Union, provisioned throughput, and compliance with enterprise data-governance requirements.
The headline feature is real-time lip-synced video avatars, though Google restricts custom avatar creation to an allowlist. A library of pre-built avatars is available to all customers, and every audio and video stream carries an invisible SynthID watermark to identify AI-generated content. The model supports 97 languages with automatic language detection and can process camera feeds and screen shares alongside audio, a capability Google calls "live visual understanding."
The foundation is a native speech-to-speech architecture: the model does not convert speech to text and back again but processes and generates audio directly. This enables natural interruption recovery without losing conversation context or candidate transactions. At the same time, the model can invoke tools and APIs in the background while continuing to speak, acknowledging the request and carrying on the dialogue while the task executes asynchronously.
Google showed three use cases. The first walks through custom avatar creation in steps: system instructions, a single reference image, and a short audio sample. The second demonstrates an insurance claims agent: the user speaks and shows damage via camera while a notebook auto-populates and a team of ADK agents checks the policy and builds an appraisal package in the background. The third demo builds a real-time voice agent with the Agent Development Kit (ADK) and the Gemini Live API, no traditional STT pipeline required.
Cox Automotive is the first named customer. It has embedded the capabilities in Autotrader, an AI shopping assistant that highlights screen regions in real time and calls tools to guide car buyers through search, comparison, and purchase. The company reports that the combination of live screen highlighting and tool calls shortens user decision time.
Gemini 3.8 Live Extended Thinking remains in private preview. The Live Avatar version is available now in Gemini Enterprise with an API for immediate development. Google emphasizes that the architecture was designed for trust and transparency: custom avatars require strict enterprise verification, and all output is marked with SynthID. Developers who want to experiment can find the claims-demo code in open source.