Gemini 3.8 Live adds real-time avatars in 97 languages

Google put a face on its live voice models on September 24, releasing Gemini 3.8 Live with Live Avatar to enterprise customers. Live Avatar pairs near real-time video generation with the speech of Gemini's live dialogue models, so an agent answers with a face and a voice generated together: lip-sync, expressions and turn-taking arrive in one pass. Google announced the feature in a blog post by research scientist Shuo-yiin Chang and software engineer CJ Zheng, and made it generally available in Gemini Enterprise through US and EU endpoints the same day.

Three things separate it from a scripted avatar video. It takes visual and audio input at once, so an agent can watch a camera feed or a shared screen while the person talks. It runs tool calls in the background without stopping the conversation, and Google's demonstration shows an avatar checking a hotel guest in while the dialogue continues. It also changes language mid-conversation, adapting lip-sync and expressions across 97 languages, which Google says holds video fidelity without visual drift.

The live dialogue models themselves launched on September 15, 2026, as Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, built for voice agents and for high-complexity reasoning respectively. Live Avatar adds the visual layer. Per Unite.AI's reading of the Gemini 3.8 Audio model card, the models are based on Gemini 3 Pro, accept audio, images, video and text with a context window up to 128K tokens, and with Live Avatar output audio, video and text against a 24K-token limit. The card states that the avatars sustain a few minutes of continuous interaction rather than extended hours, and lists hallucinations and occasional slowness or timeouts among the known limitations.

Avatars come from a preset library, or an organization can generate a custom one from a high-quality reference image that keeps the likeness, brand styling or character identity of the reference. Google says custom avatar creation is available only through enterprise allowlisting.

Customers named in Google Cloud's announcement include Cox Automotive, which is building an Autotrader shopping assistant that highlights items on screen while a shopper describes what they want. Equal AI chief executive Akhilesh Damaraju said the company's personal AI handles more than a million live calls daily across nine Indian languages, and Bob Van Osten, Salesforce's vice president of product for Agentforce, said Agentforce and Gemini 3.8 Live are being combined in a collaboration between Salesforce AI Research and Google.

Google has not published per-minute pricing for Live Avatar. Its September 15 developer post lists the underlying live models at $0.005 per minute of audio input and $0.018 per minute of audio output, and Google Cloud directs customers to its sales team for provisioned throughput and custom-avatar allowlisting.

Live Avatar is available now in the Gemini Enterprise console and through the Gemini Live API for US and EU endpoints; custom avatars require an allowlisting request.