The Gemini 3.5 Live Translate API,
Explained

Gemini 3.5 Live Translate is a real, dedicated Google model — launched in June 2026 — for near real-time speech-to-speech translation across 70+ languages. Developers reach it as the model gemini-3.5-live-translate-preview through the Gemini Live API, and Google also offers Live Translate through the Google Translate app and Google Meet speech translation preview for eligible users. What there isn't is a ready-made "Live Translate API" you simply flip on: using the model through the API still means building the app around it.

Here's the part most search results gloss over: an API is a building block, not a finished tool. If you're a developer, that distinction is exciting. If you just need to understand a client in Tokyo during a live call this afternoon, it can be a three-week detour.

You've probably noticed how fast Google's Gemini models move — new versions ship, capabilities expand, and the documentation changes every quarter. This guide explains what the Gemini Live API really does for translation, what it costs to build with, and when you're better off with a finished, browser-based real-time meeting translation tool that needs no code at all.

What is the "Gemini 3.5 Live Translate API"?

Let's clear up the name first. People search for a "Gemini 3.5 Live Translate API" expecting a single switch that turns Gemini into a live interpreter. The model is real and named — but as an API it's something you build with, not a one-click product.

You reach Gemini 3.5 Live Translate through the Gemini Live API, documented by Google for developers. It's a low-latency interface that streams audio in and streams responses back — text, and optionally synthesized speech. Live Translate is the dedicated audio model tuned for this job, while the broader Live API can also handle transcription, summarization, or question-answering with other models.

The "3.5" part is a model version. Gemini has moved through several generations, and Google lists the current options on its models page. Version branding shifts quickly, so the practical advice is simple: build against whichever Gemini model is current for the Live API, not a specific number you read in a blog headline six months ago.

The core idea: as an API, Live Translate is a model you build around, not a finished product you press play on. You open a streaming session with the speech-to-speech model, feed it audio, and it streams translated speech and text back. That flexibility is the appeal for builders — and the friction for everyone else.

Want to skip the build entirely and just read live translation in your browser? You can try MirrorCaption free with one hour of hosted transcription, no credit card. We'll come back to that — first, here's how the API path works.

How developers use the Gemini Live API for translation

At a technical level, a live translation app built on the Gemini Live API follows a clear shape. Each step is straightforward on its own; the work is in wiring them together reliably.

1. Open a streaming session. Your app authenticates with an API key and opens a persistent connection to the Live API.
2. Capture and stream audio. You grab microphone (or meeting) audio in the browser or your backend and stream it as it arrives.
3. Set the instruction. A system prompt tells the model what to do — for example, "Transcribe the speaker and translate into English."
4. Receive partial results. The API streams back text as the speaker talks, so you can display translation before the sentence ends.
5. Optionally speak it back. The Live API can return synthesized audio, so your app can play the translated voice aloud.

This is genuinely powerful. A team building a custom call-center product, a niche kiosk, or an embedded device gets a flexible engine they can shape exactly to their workflow.

Here's an illustrative example. Picture Priya, a developer prototyping a bilingual support widget. In an afternoon she has a demo that streams her voice to the Gemini Live API and prints English translation on screen. It feels like magic — until she starts listing what's left before real users can touch it: audio reconnection logic, handling silence and overlapping speakers, rate limits, billing alerts, a fallback when the connection drops mid-call, and a privacy review. The prototype was the easy 20%.

None of that is a knock on the API. It's just the honest difference between a demo and a product. The same gap is why many people who want accurate real-time translation reach for a finished tool instead of a raw model endpoint.

What building with the Gemini API really costs

The headline cost is the API bill. Google charges for the Gemini API on a usage basis, so the price scales with the model you pick and how much audio and text you push through it. That's reasonable and transparent for developers.

But the API bill is rarely the real cost. Building and maintaining a production translation feature also means:

• Engineering time to build the audio pipeline, the UI, and the error handling.
• Maintenance as model versions change and endpoints evolve — that quarterly doc churn lands on your team.
• Infrastructure for streaming, scaling, and monitoring once more than a few people use it.
• Compliance work: where audio goes, what's logged, and how you tell users.

Another illustrative example. Marco is a freelance consultant with two Mandarin-speaking clients. He doesn't need an API — he needs to understand a video call on Thursday. Pointing him at the Gemini Live API would mean hiring a developer to build something he'll use for two hours a month. The math never works. Marco wants the outcome, not the building block.

That split — builders versus everyone else — is the real decision point. So let's make it concrete.

API vs. finished tool: which do you actually need?

If you're choosing between building on the Gemini Live API and using a ready-made translation tool, this table lays out the trade-off. There's no wrong answer — it depends on whether you want a building block or a result.

If you're shipping a custom product where translation is one ingredient among many, the API is the right tool. If translation is the job, a finished tool wins on every line that isn't "infinite customization."

MirrorCaption: real-time translation without writing code

MirrorCaption is the finished-tool side of that table. It's a browser-based real-time transcription and translation app that supports 50+ selectable languages, with nothing to install and no meeting bot to approve.

For online meetings, Meet mode captures your meeting-tab audio in desktop Chrome or Microsoft Edge. No bot joins the call, so there's no awkward "recording assistant" sitting in the participant list. It works alongside browser-based Zoom, Teams, Meet, and Webex — pick whichever your host chose.

For face-to-face conversations, Talk mode turns your phone into a continuous interpreter session. You start it once and both people speak in turns; it isn't a tap-and-wait phrasebook. And with Speak Translations, MirrorCaption can read your translated speech aloud in the other language — so the person across the table hears the message, not just sees a caption.

A few details that matter for the people in Priya and Marco's shoes:

• No code, no API key. The hosted transcription and translation layer is managed for you.
• Privacy-first. Meeting audio streams through your browser and isn't stored on our servers — only the transcripts you choose to save, kept locally.
• AI summaries. A running summary refreshes as the meeting progresses, so a late joiner can catch up in one read.
• Simple pricing. A free hour to try (one-time, no monthly reset), Pro Yearly at €54.99 with 100 hours of hosted transcription included, or a €99 one-time lifetime plan with 200 hours included and all future updates. Extra hours come from Voice Packs, and lifetime customers get the lowest per-hour rate.

It's the difference between renting an engine and driving a finished car. If you're weighing other options too, our OpenAI Whisper comparison covers the same build-vs-use question for speech-to-text, and the Google Meet translation alternative page covers in-platform captions.

Three ways people use live translation today

These short, illustrative scenarios show where a finished tool earns its place over a raw API.

Cross-border sales calls

A founder in Berlin takes a call with a prospect in São Paulo. Meet mode shows the conversation in German on her screen while the prospect speaks Portuguese, and Speak Translations voices her replies back. No developer was involved — she opened a tab before the call.

Doctor visits while traveling

A traveler in Japan hands their phone across the desk at a clinic. Talk mode keeps one continuous session running as the conversation goes back and forth, so nobody has to pause and press a button for each sentence.

Multilingual standups

A distributed team runs a daily standup where the Shanghai engineer, the Berlin PM, and the Lisbon designer each read the meeting in their own language — during the call, not in a transcript emailed afterward. You can see the wider pattern in our live captions guide.

Frequently Asked Questions

Is there a Gemini 3.5 Live Translate API?

Yes — Gemini 3.5 Live Translate is a real, dedicated Google model (model code gemini-3.5-live-translate-preview) for real-time speech-to-speech translation. Developers reach it through the Gemini Live API; there's just no ready-made "API product" you switch on without building an app. Google also offers Live Translate through the Google Translate app and Google Meet speech translation preview.

Can Gemini translate speech in real time?

Yes. The Gemini Live API supports low-latency, bidirectional audio streaming, so you can stream microphone audio in and receive translated text or spoken audio back while someone is still talking — as long as you build the surrounding application.

Do I need to code to use the Gemini Live API?

Yes. It's a developer interface. You need to write code, manage an API key, capture and stream audio, and handle errors and scaling. Non-developers who just want the translated result are better served by a finished tool.

How much does the Gemini Live API cost?

The Gemini API is billed on usage, so the price depends on the model and how much audio and text you process. Google publishes current rates on its pricing page. Budget engineering time too — the API itself is only part of the cost.

What's the easiest way to translate a live meeting without an API?

Use a finished browser tool. MirrorCaption runs in desktop Chrome or Edge, captures meeting-tab audio with no bot joining the call, and shows real-time translation in 50+ languages. There's nothing to code and no API key to manage.

The bottom line

If you searched for a "Gemini 3.5 Live Translate API," the honest answer is that the model is real — but reaching it through the API gives you a powerful building block, not a finished product. It can absolutely power live translation, for teams with the engineering time to build, maintain, and scale around it.

For everyone else — the consultant with two clients, the traveler at a clinic, the team that just wants to understand each other on today's call — the building block is overkill. You want the result, and you want it now. That's exactly what a browser-based tool delivers: open a tab, pick your languages, and read every word in your own language while the conversation is still happening.

Builders, go build. Everyone else, skip the code and start translating.