Google rolled out Gemini 3.5 Live Translate on Tuesday, an audio model that starts generating translated speech before the speaker finishes talking. It covers more than 70 languages and, per the company blog, holds the trade-off between waiting for context and staying in sync, landing a few seconds behind the speaker.
The trick is prosody. The model keeps intonation, pacing, and pitch intact, so the output sounds less like flat synthesized audio. Google calls it "more seamless" translation, which is the kind of phrase that means little until you hear it. Whether it actually holds up across word orders that stack meaning at the end, like German or Japanese, isn't something the announcement proves.
It shipped straight into products. The Google Translate app gets it now on Android and iOS, with an Android listening mode that pipes translation through the earpiece like a phone call. Developers can build on it through the Live API and AI Studio in public preview.
Google Meet is the bigger play for enterprises. The blog says Meet will support 70+ languages and over 2,000 language combinations in a single meeting, up from a previous cap of five languages and English-only pairing. That's in private preview for select Workspace customers this month, broader rollout later this year.
All generated audio carries a SynthID watermark. The model card also flags limits Google's marketing skips: voices can drift, change gender, or get stuck during fast multi-speaker exchanges. Pricing for the API wasn't in the official post.
Bottom Line
Meet jumps from five supported languages to 70+ with 2,000+ combinations per meeting, now in private preview for select Workspace customers.
Quick Facts
- Model: Gemini 3.5 Live Translate (gemini-3.5-live-translate-preview)
- Languages: 70+, with 2,000+ combinations in Meet
- Announced: June 9, 2026
- Available: Google Translate (Android/iOS), Live API and AI Studio (public preview), Meet (private preview)
- All output watermarked with SynthID
