Google Debuts Gemini 3.8 Live Audio Models

Google has introduced two new real-time voice models, Gemini 3.8 Live and a heavier 'Extended Thinking' variant, promising top-tier performance at competitive pricing along with broad multilingual support.

Google is rolling out a fresh pair of voice-focused AI models aimed at real-time conversation: Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking. The announcement, shared by Google's Logan Kilpatrick, frames both as state-of-the-art live audio systems that combine strong performance with what the company calls frontier-level pricing, suggesting Google is pushing to make advanced voice AI more affordable to deploy at scale. A standout feature of the base 3.8 Live model is its language coverage — the system reportedly handles 97 languages and can switch between them mid-conversation without interruption. For developers building voice assistants or multilingual customer-facing tools, that kind of on-the-fly language switching could reduce the need for separate language-specific pipelines. The models also add support for asynchronous tool calls, a capability that lets the AI trigger external functions or fetch data without freezing up the live conversation while it waits for a response. That's a meaningful detail for anyone building voice agents that need to check a calendar, pull information from a database, or take other actions mid-chat while still sounding natural and responsive. The Extended Thinking variant appears positioned as a more deliberate version of the same live model, though Google's announcement did not detail exactly how its reasoning process differs from the standard release. Full specifics on pricing, availability, and benchmarks were shared in Google's original post announcing the launch.