Google has released Gemini 3.5 Live Translate, pushing real-time interpretation from "wait for you to finish, then translate" to "translate while you're still speaking." It supports live interpretation across more than 70 languages, with only a few seconds of latency, while preserving your speaking pace and tone.

What Gemini 3.5 Live Translate Is

Traditional translation tools work like walkie-talkies: you finish a sentence while they hold it, then they translate for the other side. Back and forth, the rhythm falls apart.

Gemini 3.5 Live Translate changes that model. It translates as it listens — the translation arrives before the speaker finishes. The model weighs, sentence by sentence, between "wait a little longer for a more accurate translation" and "start speaking now to keep up with the speaker," and the final output flows without stalls, trailing the speaker by only a few seconds.

Core Capabilities

• Translate while listening: no more waiting for you to finish before translating; the translation streams out in real time
• Preserved voice characteristics: the translated voice keeps your pace, pitch, and tone — not a cold machine voice
• 70+ language support: fully automatic detection, and it keeps up even if you switch languages mid-sentence
• Noise resistance: works in wet markets, airports, and by the roadside

Built on Gemini 3 Pro, the model handles up to 128K tokens of audio context, and the evaluation focused on three metrics: translation quality, latency, and speech naturalness.

How to Use It

Three product lines roll out at once, covering different user groups:

Consumers: Google Translate

Google Translate's Android and iOS versions are now live worldwide.

Open "Live Translate" at the bottom-left corner of the app, plug in any pair of earbuds, and go. Android also gets a "Listening mode": hold the phone to your ear like a call, and the translated audio comes straight out of the earpiece, inaudible to people around you.

Businesses: Google Meet

Starting this month, Google Meet enters private preview. It used to support 5 languages; now it jumps straight to 70+, and a single meeting can carry more than 2,000 language combinations.

Developers: Gemini Live API

Available in public beta through the Gemini Live API and Google AI Studio — you can pick it up today. Platforms like Agora, Fishjam, and LiveKit have already integrated it, wrapping up the real-time media-streaming infrastructure end to end.

Speech translation is one of Google's longest-running machine learning projects, and this time it has finally run all the way into your earbuds. -- Jeff Dean, Chief Scientist, Google DeepMind

Known Limitations

For now it accepts audio input only. Heavy accents, rapid back-and-forth language switching, several people talking over each other, or long pauses can still destabilize the voice replication.

Real-World Scenarios

Southeast Asian ride-hailing platform Grab is already trialing it. Drivers speak the local language; passengers hear their own mother tongue. Grab users place more than 10 million voice calls a month — this isn't a launch-day demo, it's a real business scenario.

Where it fits: cross-language meetings, communication while traveling abroad, multilingual customer service, online classrooms, video dubbing, and multilingual livestreams.

How to Get It

• Google Translate: Android/iOS app, "Live Translate" in the bottom-left corner
• Google Meet: private preview this month
• Gemini Live API: Google AI Studio public beta
• Model card: Gemini 3.5 Audio Model Card