Google Makes Gemini 3.8 Live with Live Avatar Generally Available to Businesses: Talking AI Avatars Come to Customer Service

What happened: Live Avatar comes to the enterprise edition

The Google Cloud Blog announced that Gemini 3.8 Live with Live Avatar is now generally available in Gemini Enterprise, Google's AI service for businesses. Generally available means the product is officially released for regular use rather than offered as a test version. Google says the technology is ready for businesses to use in their live services. That Google Cloud Blog post is dated September 25, 2026. A separate post introducing the feature was published on blog.google and is dated September 24, 2026. According to Google, the technology was first previewed at Google Cloud Next 2026.

This launch builds on Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, which Google announced on September 15, 2026. According to Google, Gemini 3.8 Live focuses on scale and cost efficiency. Extended Thinking is aimed at highly complex tasks that need several steps of reasoning. All of this information comes from Google's own websites and has not yet been confirmed by independent sources.

Read the full description

Sources are collected, independently checked, then reviewed by Jev.

Key features and current status compared

According to Google, Live Avatar can generate lip-synced avatars for conversational video agents on the web, on mobile devices and at interactive kiosks. A conversational video agent is an AI assistant that talks with people on screen. Gemini 3.8 Live understands and speaks 97 languages and can detect the language automatically, and the avatar's lip movements and expressions can switch across these languages. The model can also make tool and API calls in the background while the conversation continues. In plain terms, it can look things up or trigger actions in other software without pausing the chat. Its real-time visual understanding can process camera feeds, screen sharing and audio at the same time. Google also says the generally available version offers US and EU endpoints (server access points in those regions) and provisioned throughput (reserved processing capacity). It also includes enterprise compliance and data governance.

On performance, Google points to results on industry benchmarks, which are standardized tests used to compare AI models. Google says Gemini 3.8 Live Extended Thinking ranks first on the Artificial Analysis speech-to-speech quality index with a score of 82.6. It also scores 68.6% on τ-Voice and 97.7% on Big Bench Audio. Google says Gemini 3.8 Live took second place on the Speech Agent Arena. Google published these results itself, so readers should treat them as vendor claims.

Identity protection and watermarking

AI avatars that can speak in real time with a human likeness raise two main concerns: impersonation and deception. According to Google, customers can choose from a library of prebuilt avatars. To generate a custom avatar from reference images, a business must first be placed on an allowlist, an approved list of customers, and pass a verification process. Google also says all generated audio and video streams carry SynthID watermarks. SynthID is Google's hidden digital marker, which people cannot notice by eye or ear. It is intended to make AI-generated content detectable.

What it means for everyday readers

This release is aimed at businesses and developers, so consumers will not buy or switch on the service directly. In practice, businesses that adopt it may put AI avatars into customer service, shopping help or check-in processes. The Google Cloud Blog gives the example of Cox Automotive, which built an AI shopping assistant for Autotrader that helps car buyers search for and compare vehicles. Google also showed an insurance claims demo in which a user describes damage to the camera. Meanwhile, an AI system working in the background checks the policy and organizes the information.

The Google Cloud Blog also quotes Equal AI CEO Akhilesh Damaraju, who says its AI handles more than one million live calls a day in nine Indian languages. It also quotes Salesforce's Bob Van Osten on the collaboration between Gemini 3.8 Live and Agentforce. Google selected these customer statements. Multilingual AI customer service that can see what you show it may make some services more convenient. It also means a company's AI system will process the images and audio you share with the camera, so check that company's privacy notice before using it.

FAQ

Browse the latest news in this topic

• Tech news