Google's new voice model switches between 97 languages on its own.
Google released Gemini 3.8 Live on Tuesday, a model that hears speech and answers in speech. It notices the language you switch to and follows, across 97 of them, without being told. The heavier version took first place on an independent speech-to-speech ranking.
Why it mattersVoice is where AI stops needing a keyboard, and language switching is where it usually breaks. This model already sits inside Search and the Gemini app.
▶ Related video: Gemini 3.8 Live is not a voice model, it's an agent
Read in the app — free, in 9 languages
Related stories
Alibaba cut the cost of feeding an hour of audio to AI by 98%.2026-09-19Salesforce trained its own reasoning model so your data stays put.
2026-09-15Shanghai's top open agent model arrived with no announcement at all.
2026-09-15DeepSeek is testing its first voice mode on a slice of users.
2026-09-14Xiaomi's open model picks one voice out of three talking at once.
2026-09-14