Gemini 3.5 Transcribe: Faster Dictation, With an Editing Risk

Google’s new Gemini 3.5 Transcribe makes voice input faster and cleaner by removing filler words and spoken corrections, with a reported 5.5% live-speech error rate. It is available in selected Google products and through the Gemini API, with Chrome promised soon. For your business, test it on routine drafting, but keep exact-wording work out of the first trial.

What does Gemini 3.5 Transcribe actually do?

This is speech-to-text with an editorial layer. As well as recognising words, Gemini 3.5 Transcribe can remove verbal clutter such as “um” and “uh”, apply corrections you make while speaking and use custom vocabulary for specialised terms.

Google says the model works across 85 languages. It can also process pre-recorded audio containing as many as three speakers, although the supplied report does not describe how those speakers appear in the finished transcript.

How much faster and more accurate is it?

Google reports that Gemini 3.5 Transcribe cuts the time from speech to finished text by about 70% compared with Chirp 3, its previous voice-to-text engine. That could make dictation feel less like waiting for a slow typist to catch up.

The reported live-speech error rate is 5.5%, down from 7.32% for Chirp 3. That is an improvement, but it is not a promise of flawless copy, particularly when the system is doing more than simply recording the words it hears.

For an operator, the likely saving is small but frequent: less time deleting filler words, repairing false starts and cleaning routine drafts. It replaces part of the manual tidy-up after dictation, while Chirp 3 is the named engine it directly succeeds within Google’s system.

Where can your business use it?

Gemini 3.5 Transcribe already powers Rambler, a Gboard feature currently limited to Pixel 11 phones. Google plans to extend Rambler to more devices with Gemini Intelligence later in 2026, according to the reported announcement.

The model also became available for voice input in the Gemini app on macOS on August 26, 2026. Developers can access it through the Gemini API, while Google’s Antigravity and AI Studio build model also received support.

Chrome support is due “soon”, without a firm date in the source. Once released, Google says the feature will accept voice input in any web text field, which would make it useful for drafting emails, filling forms and writing prompts without changing applications.

Should you switch now?

There is no published price in the supplied material, so a proper cost comparison is not yet possible. Availability is also uneven: some developers and macOS users can start now, while many phone and Chrome users must wait.

The larger issue is control. Gemini 3.5 Transcribe is designed to capture your intended meaning, not preserve every word exactly, and its clean-up process can alter your phrasing. That trade can be helpful for a quick draft and unsuitable when the original wording is the record you need.

Start with a narrow trial using routine internal drafts. Give it your specialised vocabulary, compare several outputs with the original speech and note how often someone must repair the result. If the clean-up saves more time than checking consumes, expand its use; if wording fidelity matters, keep conventional transcription in place.