Gemini 3.5 Live Translate Brings Near Real-Time Speech to Meet and Translate

Google DeepMind has released Gemini 3.5 Live Translate, a model built to handle near real-time voice translation with natural-sounding speech. The model is rolling out across three Google products: Google AI Studio, Google Translate, and Google Meet. The focus, according to Google DeepMind, is on fluid, natural output rather than the stilted, word-by-word style that has made earlier real-time translation tools frustrating to use in actual conversations.

What happened

Google DeepMind announced Gemini 3.5 Live Translate and confirmed it is being deployed in three products simultaneously: Google AI Studio, Google Translate, and Google Meet.

The stated goal is near real-time translation that sounds natural. That means the model is not just converting words one at a time. It is trying to preserve the rhythm, tone, and flow of spoken language across the translation.

Google AI Studio gives developers direct access to test and build on the capability. Google Translate and Google Meet bring it to a much wider base of everyday users, including anyone using Meet for cross-language business calls.

Why it matters

Real-time voice translation has been technically possible for a while, but the quality gap has kept it from being genuinely useful. Most live translation tools produce output that sounds mechanical, lags noticeably, or loses meaning in ways that derail conversations.

Google is making a specific claim here: that Gemini 3.5 Live Translate closes that quality gap enough to make the experience feel fluid. If that holds up in practice, it changes how teams with multilingual members or international clients can use Meet.

For businesses already paying for Google Workspace, getting a usable live translation layer inside Meet would be a meaningful addition at no extra tool cost. That is worth watching.

The AI Studio access also matters for developers. Being able to build near real-time voice translation into a product or workflow without training your own model is a significant capability to have available through an API.

Our take

Google has a long history of promising natural-sounding translation and then shipping something that still sounds like a robot reading a dictionary. That pattern is worth keeping in mind here.

That said, Gemini 3.5 is a capable model family, and the architecture behind live audio processing has improved sharply over the past two years. The combination of low latency and natural prosody is genuinely hard to get right, so it is worth testing rather than dismissing.

A few things we would want to know before recommending this to a client for live business calls:

• Which languages are supported, and how does quality vary between high-resource and low-resource language pairs?
• How does it handle heavy accents, fast speech, or domain-specific vocabulary?
• What are the latency numbers in real network conditions, not a demo environment?

Until those answers are clear, treat this as a promising beta feature rather than a replacement for a human interpreter on a high-stakes call.

What to do about it

If your team uses Google Meet for calls with non-English speakers, test the Live Translate feature on a low-stakes internal call first. If you are a developer, open Google AI Studio and run the model against a real audio sample from your use case. That will tell you more than any announcement post.

Source: Google DeepMind