Key Takeaways
- Launch: On September 24, 2026, Google announced Gemini 3.8 Live with Live Avatar, now in general availability for Gemini Enterprise customers in the United States and the European Union.
- Technology: an animated avatar with near-real-time video generation, lip-sync across 97 languages with mid-sentence language switching, and asynchronous background tool execution.
- Adoption: early live customers include Cox Automotive, Equal AI, and Salesforce, in a market where Google is chasing rivals like HeyGen, OpenAI, and Anthropic.
A natively multimodal dialogue
Google has paired near-real-time video generation with its own native voice dialogue models. The avatar processes visual and audio input simultaneously, looks at what the user shows via camera or screen share, and responds with facial expressions, lip-sync, and fluid conversation turns. It's not a preloaded video bolted onto a voice synthesizer: it's visual presence generated in the moment.
The feature is available for Gemini Enterprise customers, with endpoints in the United States and the European Union, dedicated throughput, and enterprise compliance.
97 languages, zero interruptions
Live Avatar supports 97 languages and can switch between them mid-conversation without degrading video fidelity or introducing visual drift. According to The Verge, a demo video shows the avatar speaking English and Japanese with mouth animation aligned to both languages. For global businesses, this means a single visual agent for every market, with automatic language detection and no manual switches.
Works while it talks
The avatar can trigger system calls, verify data, and complete complex tasks in the background while the conversation continues uninterrupted. In one Google demo, the avatar checks in a hotel guest while chatting with them as system operations run in parallel. In another demonstration, an insurance agent analyzes damage shown via the customer's camera while a team of agents built with the Agent Development Kit verifies the policy and prepares the claims file.
Developers can integrate these features through Google's Agent Development Kit, defining memory and audio streaming without going through traditional speech-to-text pipelines.
A face tailored to the brand
Organizations can choose from a library of preset avatars or generate custom ones starting from a reference image, preserving likeness, brand style, or character identity. Creating custom avatars requires access via enterprise allowlist, a filter Google put in place deliberately.
The trust question: SynthID and safeguards
Google states it has put safeguards in place to respect identity and keep AI-generated content transparent, with an invisible SynthID watermark stamped on all generated audio and video streams.
Who's already building on it
Cox Automotive built a shopping assistant for Autotrader that guides buyers through researching and comparing vehicles in real time. Equal AI handles over a million live calls a day across nine Indian languages, reporting improvements in handling interruptions and multilingual conversations. Salesforce is integrating Agentforce with Gemini 3.8 Live's capabilities.
What it means for everyone else
Live Avatar variants currently sustain only a few minutes of continuous interaction, not hours, and for now the feature remains enterprise-only. The launch comes at a delicate moment for Google, amid reported delays on Gemini 3.5 Pro and competing launches from OpenAI and Anthropic. Live Avatar pits Google against startups like HeyGen, with one advantage: the avatars aren't a separate product, but built directly into the model's runtime.
