The dream of talking to anyone in any language in real time has been promised so many times it became a cliché. Google thinks this time it’s actually solved it.
Google launched Gemini 3.5 Live Translate — a new AI model for real-time speech translation that supports over 70 languages and opens up more than 2,000 language combinations in a single conversation. It’s rolling out now in Google Translate on Android and iOS.
The part that makes this different from every previous attempt is how it handles timing. Traditional translation systems wait for the speaker to finish a sentence before responding — creating those awkward silences that make real-time conversation feel stilted and unnatural. Gemini 3.5 Live Translate generates speech continuously, staying only a few seconds behind the speaker throughout the entire session. The result is something much closer to how human interpreters actually work — overlapping, anticipating, keeping pace.
What you’ll actually see in the apps
In Google Translate, you activate it by tapping “Live translate” in the bottom-left corner while wearing headphones. For Android, there’s also a new listening mode that lets you hold the phone to your ear like a normal phone call — no headphones required, just the regular speaker. Low-friction enough that you’d actually use it at an airport, in a meeting, or on the street.
The bigger deal is what’s happening in Google Meet. Live translation previously supported only five languages. Gemini 3.5 Live Translate expands that to over 70, unlocking more than 2,000 language combinations — and critically, it breaks out of the old English-only pivot model where translation only worked to and from English. Two people speaking Spanish and Japanese can now have a Meet call without either of them reaching for a workaround.
The catch
The Meet integration is rolling out as a closed preview for select Google Workspace business customers this month. A broader launch is planned for later in the year. Consumer users on free accounts will likely be waiting a while.
On the developer side, the model is already available in public preview via the Gemini Live API and Google AI Studio — meaning this technology is about to show up in a lot of third-party apps well before the full Workspace rollout. Multilingual customer service tools, online tutoring platforms, international live streams — all of them now have a real-time translation backbone to build on that didn’t exist last week.
The privacy piece
Google says all audio generated by the model is marked using SynthID, an imperceptible watermark woven directly into the audio output that identifies it as AI-generated content — a mechanism designed to help prevent the translated audio from being used to spread misinformation. It’s a small but notable detail: the company is building detection capability into the output from day one, rather than scrambling to add it later.
The model also handles real-world conditions better than previous versions — it can automatically detect which language the other person is speaking and maintains stable performance even in noisy environments, which matters enormously when the use case is an actual conversation rather than a controlled demo.
Real-time translation at this quality and this scale would have been a research moonshot five years ago. Today Google is pushing it as a Translate update.