Gemini 3.5 Live Translate pushes real-time voice translation beyond 70 languages

Real-time voice translation is moving from a demo feature into a model layer that developers and meeting products can call directly. Google DeepMind wrote in a June 9 tweet that Gemini 3.5 Live Translate is built for “fast, cross-language communication,” positioning it as the latest audio model in the Gemini line.

The concrete scope is more than 70 languages. Follow-up material from Google says the model translates speech as it is streamed and aims to preserve tone, pace, and pitch. The access paths are split by audience: Google Translate on Android and iOS for general users, public preview through the Gemini Live API and Google AI Studio for developers, and private preview in Google Meet for enterprises.

Google DeepMind’s account usually mixes research posts, Gemini model updates, and product deployments across Google. This one matters because the distribution is broad. If translation happens continuously during speech rather than after each completed sentence, customer support, remote collaboration, education, and travel apps can redesign around live multilingual conversation instead of turn-taking translation boxes.

The next watch item is measured performance outside Google’s examples. More than 70 languages is a large coverage claim, but production use will depend on latency, background-noise tolerance, terminology handling, and whether preserved voice characteristics remain natural across accents and language pairs. Because the developer preview is available through Google AI Studio and the Gemini Live API, independent tests should appear quickly.