Google Just Built a Translator That Keeps Up With You in Real Time. Here’s What’s Actually Different.

A quick recap: what was wrong with translation before?

If you have ever used a live translator in a meeting or a phone call, you know the pain. You say something. The system waits for you to finish. Then it translates. Then the other person waits. Then they respond. Then it translates again.

It feels like talking through a broken walkie-talkie. The pauses are awkward. The rhythm is lost. And by the time the translation comes through, the moment has already passed.

Every major translation tool — including older versions of Google Translate — worked this way. Turn by turn. Wait, translate, wait, translate. It was functional. It was not fluid.

On June 9, 2026, Google changed that.

So what is Gemini 3.5 Live Translate?

Gemini 3.5 Live Translate is Google’s new audio model for live, speech-to-speech translation. It supports over 70 languages and it does not wait for you to finish speaking before it starts translating.

Instead it processes speech as you stream it — continuously, in near real time — staying just a few seconds behind the speaker throughout the entire conversation.

Here is what makes it genuinely different from what came before:

It does not make you wait — Traditional systems translate turn by turn. This one generates translated speech continuously, so the conversation flows like a real conversation rather than a structured relay.

It preserves how you sound — The model captures your intonation, your pacing, your pitch. The translated version does not sound like a robot reading your words. It sounds like you, speaking another language.

It handles noise — Loud environments, background chatter, unpredictable audio — the model is built to be robust in the kinds of real-world situations where you actually need a translator.

It auto-detects languages — You do not need to configure settings or tell it what language is being spoken. It figures that out on its own, across 70+ languages.

Where can you actually use it?

This is not a research announcement. It is rolling out right now across three places:

Google Translate app — Available today on both Android and iOS. Connect any pair of headphones and the Live Translate feature uses the new model. Android users also get a new “listening mode” — hold your phone to your ear like a regular call and the translated audio streams directly through your earpiece. No headphones needed, and no one else hears it.

 

Google Meet — Rolling out in private preview for business Workspace customers this month, with a broader rollout later this year. The upgrade is significant: Meet previously supported only five languages for speech translation and could only translate to and from English. With 3.5 Live Translate, it supports 70+ languages and over 2,000 language combinations in a single meeting.

Gemini Live API and Google AI Studio — Available now in public preview for developers who want to build their own voice translation apps on top of the model.

Who is already using it?

Grab — the Southeast Asian ride-hailing giant has been testing 3.5 Live Translate to enable real-time communication between drivers and passengers who do not share a language. Their users make over 10 million voice calls per month through the app. That is not a small pilot.

Developer platforms like Agora, LiveKit, and Pipecat are already integrating the Gemini Live API, making it easier to build and deploy voice translation apps without having to build the real-time streaming infrastructure from scratch.

What is the invisible thing happening here?

Here is what the announcement does not say loudly enough.

Twenty years ago, Google launched machine translation as an experiment. Today, over a trillion words are translated every month across Google’s products. That is the scale this model is now plugging into.

But the more important shift is not the scale. It is the paradigm. Turn-by-turn translation was always a workaround — a way of making the technology fit inside the constraints of what AI could do at the time. Real-time, continuous, voice-preserving translation is what people actually needed. It just was not possible until now.

There is also something quietly significant about the SynthID watermarking. Every piece of audio this model generates is watermarked at the signal level — imperceptibly, but detectably. In a world where AI-generated audio is increasingly hard to identify, Google is building the accountability layer directly into the output. That matters more than it might seem right now.

What is happening now?

As of June 9, 2026, Gemini 3.5 Live Translate is live in the Google Translate app on Android and iOS globally.

Google Meet’s private preview for enterprise Workspace customers begins this month, with a public rollout expected later in 2026.

Developers can access the model today via the Gemini Live API and Google AI Studio.

The gap between speaking and being understood just got a lot smaller.

Source including images: Google — The Keyword Blog, Grab Newsroom