Google Meet Live Translation Update: 70+ Languages With Gemini 3.5

Google is rolling out Gemini 3.5 Live Translate today across three surfaces at once: the Google Translate app globally on Android and iOS, Google Meet for enterprise customers entering private preview, and the Gemini Live API for developers in public preview, according to Google's announcement. The Meet piece carries the biggest numbers: Google says the feature now supports 70+ languages and more than 2,000 direct language combinations in a single call, up from four languages that routed everything through English nine months ago.

That jump from four languages to 70+ is not a gradual expansion. It's a architectural change in what the feature can do.

What's available today, and what isn't

The rollout splits cleanly by audience, and the differences matter for anyone planning around it.

Google Translate live translation (consumer, global): Gemini 3.5 Live Translate is available today on Android and iOS worldwide, with no sign-up or preview gate, Google confirmed. Six months ago, the live translation beta was Android-only and limited to the U.S., Mexico, and India. Google said at the time it would bring the feature to iOS and more countries in 2026. Today's release delivers on that.

Google Meet (enterprise, private preview): The 70+ language expansion enters private preview this month for enterprise customers, Google noted. Private preview means limited access and no broad customer commitments yet. General availability timing, pricing, and Workspace tier requirements have not been announced. That last gap is significant: organizations evaluating the feature have no way to assess licensing costs or whether their current plan will include it.

Gemini Live API and Google AI Studio (developers, public preview): Third-party developers can access Gemini 3.5 Live Translate in public preview starting today, per Google. Public preview means broader access than the enterprise track, but it's still pre-general availability.

The contextual translation features Google added to the Translate app in February 2026, including tone-matching, idiom alternatives, and register guidance, remain available only in the U.S. and India, Google noted at the time. Not all Gemini-powered Translate features share the same geographic footprint, and today's global rollout doesn't change that.

Google Meet speech translation: what changed under the hood

The technical shift matters more than the language count, because it determines whether real-time voice translation can actually support a conversation or just technically exist in one.

Older audio translation systems worked in sequence: transcribe speech to text, translate the text, then convert the result back to audio. That three-step chain produced latency of 10 to 20 seconds and generated generic synthesized voices bearing no resemblance to the original speaker, as Google's engineering team described nine months ago. At a 15-second lag, you're not having a conversation; you're waiting for one to be handed back to you.

Gemini 3.5 processes speech natively, without that intermediate text step. Google says the result tracks just a few seconds behind the speaker while preserving intonation, pacing, and pitch, per the announcement. Two to three seconds turned out to be the practical target, according to Google's September 2025 engineering writeup: faster became hard to follow; slower broke the rhythm of conversation.

The language routing change is equally structural. Previously, every Meet translation passed through English as an intermediate step, meaning a Mandarin-to-Spanish exchange was actually Mandarin-to-English-to-Spanish. Google says the new system supports more than 2,000 language combinations directly, without that English bridge, per the announcement. For a call where no participant uses English at all, that's what makes the feature functional rather than nominal.

One limit worth stating plainly: all the naturalness and voice-preservation claims come from Google's own materials. No independent benchmarks or third-party testing are available as of today. Google cited leading performance on the WMT25 Machine Translation benchmark for text quality, per its December 2025 announcement, but WMT25 measures written translation, not live speech. How the model handles noisy audio, overlapping speakers in a group call, or lower-resource languages at the edges of the 70+ list has not been tested publicly. Those are open questions, not settled ones.

What the developer API access enables

Opening Gemini 3.5 Live Translate to developers through a public API is the least-reported part of today's announcement, and it carries long-term implications the consumer rollout doesn't.

When a capability enters a public API, it stops being a Google product feature and becomes something third parties can build on. Developers could embed real-time voice translation into telehealth platforms for cross-language patient consultations, legal or HR tools where accuracy carries professional weight, or consumer communication apps for live international exchange, all without building translation infrastructure themselves. The model functions as a service underneath, the same way cloud storage or speech-to-text already do for third-party builders.

Google hasn't named specific developer use cases in the announcement materials. But the structure of public preview via the Gemini Live API and Google AI Studio signals that Google is positioning translation as a layer other products draw from, not only a feature it ships directly.

This fits the pattern of the past six months. In December 2025, Google brought Gemini-powered text translation to Search and the Translate app. In February 2026, it added contextual and tone-aware features. Today's release extends live speech translation to consumers globally, enterprise customers in preview, and now to developers as a buildable primitive, per Google's published timeline. Each step has extended the capability further.

What Google hasn't addressed

Today's announcement is detailed on what it covers. Several things that will determine real-world value remain unspecified.

Translation quality across 70+ languages is not uniform, or at least Google's materials don't say it is. The 2,000+ combinations figure doesn't indicate whether all pairs perform at the same level or whether some combinations work better in one direction than the other. For enterprise customers preparing multilingual calls, that distinction matters considerably.

Pricing and access requirements for Meet are still unknown. No general availability date has been announced. Organizations interested in the feature have no way to model costs or determine whether their current Workspace plan will include it when it exits preview.

Data handling and privacy practices for translated speech in Meet and the developer API aren't addressed in available materials. That gap will matter to enterprise procurement teams, regulated industries like healthcare and legal services, and any developer building on the API in a context where client data confidentiality is a legal requirement. It's not an unusual omission for a day-one announcement, but it's one that will need answering before the feature can move into sensitive professional environments.

Independent validation is absent. The performance record consists entirely of Google's own claims. That's standard for a launch announcement, but it shapes how much weight any quality benchmark can carry right now.

The next meaningful signals will be a Meet general availability date, early developer adoption patterns through the API, and independent accuracy testing across lower-resource language pairs. Until those surface, the question of whether today's rollout holds up at scale and at the edges of the language list remains open.