Skip to content
NewsProduct launch

Gemini 3.8 Live Avatar reaches general availability in Gemini Enterprise

· by Pondero Newsdesk

The short version

Google's Gemini 3.8 Live Avatar generates a lip-synced, expressive face in the same pass as speech and switches among 97 languages unprompted, now GA in Gemini Enterprise.

Gemini 3.8 Live Avatar reaches general availability in Gemini Enterprise

Most voice bots still work in two steps: transcribe the words, then generate a reply. Google's newest release collapses that into one pass that outputs a lip-synced, expressive face alongside the speech itself, and switches among 97 languages without anyone telling it which one a caller is using.

What

Google made Gemini 3.8 Live with Live Avatar generally available on September 24, 2026, inside Gemini Enterprise on US and EU endpoints, per Google's announcement. The model runs native speech to speech, so there is no separate speech-to-text step in the middle, and Google says it produces precise lip-syncing, natural facial expressions, and fluid turn-taking rather than the pause-and-respond pattern of earlier voice agents. Tool calls run in the background while the avatar keeps talking, and a caller can interrupt mid-sentence and have the avatar recover. Every second of output audio and video carries a SynthID watermark, which the company says makes generated content detectable to reduce misinformation risk. Custom avatars built from a brand's own reference images exist but are restricted to enterprise customers who apply for allowlisting, not open to every Gemini Enterprise account.

Enterprise pricing is public, latency numbers are not

Google published a rate card running through December 31, 2026: $0.75 per million text-input tokens, $3.00 per million audio-input tokens, $12.00 per million audio-output tokens, and $1.00 per million video-output tokens, with video billed only while the avatar is actively speaking rather than while it listens, per DataNorth's review of the pricing. That works out to roughly $0.39 for one minute of the avatar speaking and about $4 for a ten-minute guided session, per DataNorth's calculation. A team evaluating a customer-service or onboarding-walkthrough deployment can model that cost today. What they cannot model yet is responsiveness: Google has not published a latency figure or time-to-first-frame number, so whether a session feels conversational or sluggish will only show up once a team tests its own workload against the live endpoint.

Context and reactions

Google frames the launch around enterprise customer service and interactive product walkthroughs rather than consumer chat, and access is currently limited to Gemini Enterprise, not the consumer Gemini app, per Google's blog post. This is a second capability layered onto the Gemini 3.8 Live line in the same week Pondero covered its Extended Thinking update, suggesting Google is iterating on the Live product faster than a single monthly release cadence.

What to watch next

Whether Live Avatar expands beyond Gemini Enterprise into the consumer Gemini app, and whether Google eventually publishes latency or time-to-first-frame benchmarks now that video generation is metered as a distinct line item from audio.

Sources