Google rolls out Gemini 3.8 Live with Live Avatar in Gemini Enterprise, lip-sync across 97 languages

Google made Gemini 3.8 Live with Live Avatar generally available in Gemini Enterprise on September 24. The release layers low-latency video generation on top of the Gemini 3.8 Live voice model Google shipped a week earlier, combining live audio and camera input with synchronous speech, lip-sync, and facial expressions. Every audio and video frame is stamped with Google's SynthID watermark. The model maintains lip-sync across 97 languages even when switching mid-conversation, outputs 24 frames-per-second video with 24 kHz voice, and supports asynchronous tool calls so the avatar can fetch data while continuing to talk.

Live Avatar supports asynchronous tool calling, letting the model answer questions and complete tasks in parallel. A hotel check-in scenario would let the avatar continue speaking with the guest while the system completes booking in the background, with no conversation pause. Google is targeting enterprise use cases including customer service, interactive walkthroughs, and virtual reception. Launch customers include Salesforce Agentforce and Cox Automotive.

Pricing bills video only while the avatar is speaking. Per million tokens, text input is $0.75, audio input is $3.00, audio output is $12.00, and video output is $1.00. The introductory rates run until December 31, 2026. Google describes Live Avatar as the first product to fold avatar generation into the same inference pass as the voice, putting direct pressure on Synthesia, HeyGen, and D-ID, whose core pitch was bolting avatars onto a model. Custom avatars built from a reference image remain gated behind an enterprise allowlist.