Our latest speech-to-text model designed for precise and intelligent real-time transcription.

I’m happy to share this! 🚀 Gemini 3.5 Transcribe is a big step forward for natural voice interaction.

Instead of typing every thought, it works the way you actually talk:

Handles self-corrections
Removes filler words & formats clean text
Understands natural intent & speaking style
Works accurately in noisy environments
Up to 3 speakers with timestamps
85+ languages, accents & dialects

On Gemini for macOS, you can use your voice to generate images, search, summarize, analyze files and more.

It also powers Rambler on Android, turning spoken thoughts into polished text with voice-based edits, corrections and style changes.

🚀 Rolling out now on Gemini for macOS and Rambler on Android. Developers can build with it in Google AI Studio and Google Antigravity.

My experience here has been different. IMO, Google needs to stop advertising new models before they shift capacity to run them. Every time I go to build something on a newly promoted Gemini model, the response times are super long, if they respond at all.

I've been using 3.5 for a year and it works really well

@haonanlin wow you must come from the future to be able to use it for a year... or just an AI without context

This works really well.. we use it almost all the time.. although Google doesn't need it.. still good wishes..

Congrats on the launch. Good to see a model that cleans up the filler rather than transcribing every word faithfully.