Gemini 3.8 Live removes the blank face from AI customer service

The system is Gemini 3.8 Live with Live Avatar. It couples Google‘s speech-to-speech model with a real-time video layer that generates lip movement and facial expressions, all while handling background tool calls without pausing the conversation. Language detection is automatic — the avatar switches between spoken languages mid-call without a handoff, a system reload, or a perceptible pause.

What separates this from video overlay techniques is that audio and video generation run in the same live inference path rather than as sequential steps. That coupling keeps the latency low enough for natural conversation rhythm. Google embeds SynthID watermarking imperceptibly into both the audio and video output — every second carries a machine-readable tag identifying it as AI-generated, even if the stream is recorded and played back later. The watermark survives re-encoding, so provenance verification remains possible on exported clips.

Equal AI, handling enterprise deployments across India, gives the clearest indication of what this enables at operational scale: nine Indian languages active simultaneously, more than one million live calls per day, across a total supported range of 97 languages. That throughput would be economically impractical with human agents at equivalent availability and scheduling coverage. The design assumption behind the deployment is not proof-of-concept — it is throughput.

What the launch withholds is the price. Google has not published costs, directing inquiries to its enterprise sales team. Custom avatar creation — organizations building a face from their own reference images — requires enterprise allowlisting through a separate verification process, not self-service. Extended Thinking capabilities for the Live model remain in private preview, capping reasoning depth compared to Google’s other Gemini offerings. Concurrent session limits have not been disclosed. The restricted launch profile is consistent with Google’s pattern for staged enterprise releases where pricing and capacity are negotiated rather than published.

Gemini 3.8 Live with Live Avatar is available now in the Gemini Enterprise console and via the Live API, with endpoints in the United States and European Union. Extended Thinking for the Live model carries no published timeline for broader access beyond the current private preview cohort.

The enterprises deploying it will answer a question the technology does not: whether removing the visual signal that marks an automated interaction improves the experience for users, or removes the last honest indicator that one is happening.

Tags: artificial intelligence, Google, Gemini, speech recognition, customer service, enterprise AI