Google Launches Gemini 3.5 Live Translate for Real-Time Voice Translation in 70+ Languages

Google announced on Tuesday, June 9, 2026, the launch of its latest audio model, Gemini 3.5 Live Translate, dedicated to real-time speech-to-speech translation across more than 70 languages. The announcement comes twenty years after Google's translation project began, a project that today translates more than a trillion words monthly for billions of users across its products.

Continuous Translation Without Waiting

The model's standout feature is that it generates translated speech continuously rather than waiting for the speaker to finish their sentence, as is the case in traditional turn-based translation systems. This approach balances waiting briefly to improve contextual accuracy with translating immediately to stay synchronized with the speaker, producing fluid audio without awkward pauses and staying just a few seconds behind the speaker throughout the session.

The model detects languages automatically without manual setup, and generates natural-sounding speech that preserves the speaker's intonation, pacing, and pitch. It is also designed to work in loud, unpredictable environments thanks to its noise robustness.

Where Is the Model Available?

Gemini 3.5 Live Translate begins rolling out today across three surfaces. It is available to developers in public preview through the Gemini Live API and Google AI Studio, under the model identifier gemini-3.5-live-translate-preview, and can be used for cases such as live dubbing of video content and real-time multilingual translation. Enterprises get a private preview within Google Meet this month, while everyone can access it through the Google Translate app on Android and iOS.

In Google Meet specifically, language support jumps from just five languages to over 70, enabling more than two thousand possible language combinations within a single meeting. In the Translate app, the live translation feature works with any connected headphones, and the Android version added a listening mode that lets you hold the phone to your ear like a regular call, so the translated audio reaches you through the earpiece without others hearing.

Technical Foundation and Safety

Gemini 3.5 Live Translate is based on the Gemini 3 Pro model, with an audio context window of up to 128K tokens. Notably, all audio the model generates carries an inaudible SynthID watermark, a measure intended to help combat misinformation and distinguish AI-generated audio from real audio.

Early Practical Use

Some companies have already begun testing the model; the ride-hailing service Grab is trying it to ease communication between drivers and passengers at pickup points, noting that its users make more than ten million voice calls per month. Other organizations reported positive feedback on translation quality, accuracy, and low latency.

What Does This Mean for Arabic Users?

With support for more than 70 languages and automatic language detection, the model opens a wider door to instant communication across language barriers in meetings, business, and travel. For developers in the region, the Gemini Live API enables building live translation and content dubbing apps without constructing a complex audio infrastructure from scratch, though measuring translation quality across diverse Arabic contexts remains something worth testing in practice.