Google has moved its multimodal voice AI agent platform from preview into general availability, adding a visual avatar layer that allows conversational agents to display a face that moves in sync with speech. The feature, called Gemini 3.8 Live with Live Avatar, is now available through Gemini Enterprise and was first previewed at Google Cloud Next 2026 earlier this year.
What Live Avatar Actually Adds
Gemini 3.8 Live already provided a native speech-to-speech foundation, meaning conversations flow without the latency and stutter that older text-to-speech pipelines introduced. The Live Avatar layer adds a visual component: conversational video agents that display synchronized lip-syncing and can operate across web browsers, mobile applications, and interactive kiosks. The agent can see what the user sees through live camera feeds and screen shares processed simultaneously with the audio stream.
Under the hood, the platform supports 97 languages with automatic language detection, executes tools and API calls in the background while continuing the conversation, and recovers naturally from interruptions without dropping context or backend transactions. The system also integrates with Google's Agent Development Kit, allowing developers to define agents, manage session memory, and stream real-time audio directly to the Gemini Live API without routing through a traditional speech-to-text intermediate step.
Trust and Verification Built In
Any technology that generates realistic video faces raises obvious concerns about identity and misuse. Google's approach addresses this through a layered verification system. Customers can deploy from a library of curated, pre-built avatars without additional overhead. Custom avatar creation is gated behind a strict enterprise allowlisting and verification process rather than being openly available.
Beyond access controls, every generated audio and video stream carries imperceptible SynthID watermarks. These marks embed identifying metadata directly into the media output, ensuring that AI-generated content remains transparent and verifiable by downstream recipients. The combination of access gating and embedded watermarking is designed to give enterprises a defensible chain of custody for the media their agents produce.
Where It Is Being Used
Several enterprise customers have already deployed the platform in production settings. Cox Automotive built an AI-powered shopping assistant for Autotrader that uses live screen-highlighting and tool-calling to guide car shoppers through vehicle search, comparison, and financing in real time through natural conversation. Equal AI, a company focused on voice-first customer service, reported that its system now handles over a million live calls daily across nine Indian languages and that the Gemini 3.8 Live improvements to interruption handling and multilingual conversations directly increased reliability.
Salesforce AI Research is integrating Gemini 3.8 Live with its Agentforce platform, combining real-time multimodal capabilities with agentic AI to explore customer experiences that span from first contact through resolution. SPECS, a platform provider, reported that the updates to Voice Activity Detection and overall latency improvements are central to the quality of its AI assistant offering.
Availability and Limits
Gemini 3.8 Live with Live Avatar is currently available through US and EU endpoints with provisioned throughput, enterprise compliance, and strict data governance. A companion feature, Gemini 3.8 Live Extended Thinking, remains in private preview and is not yet generally available.
Custom avatars are restricted to an allowlist, meaning enterprises cannot simply upload any reference photo and expect it to be approved for production use. The company has not disclosed the specific criteria for allowlist approval. Pricing details are available through the Google Cloud pricing page rather than published openly.
The general availability of a feature that combines voice, video, visual understanding, and tool execution into a single API reflects a broader shift in enterprise AI. The competitive conversation has moved past raw speed and cost per token toward how naturally an agent can interact and how much work it can complete without human intervention. Live Avatar is Google's answer to that question, packaged as a production-ready service rather than a research prototype.