Gemini 3.5 Transcribe

Two speech-to-text models from Google, one for transcribing recordings and one for streaming live audio.

gemini-3.5-transcribe handles recorded files. It detects the language per utterance across 85+ languages, labels who is speaking, and timestamps every word.
You can feed it a list of up to 1,000 custom terms so it gets product names, people, and industry jargon right.
gemini-3.5-transcribe-live streams over a WebSocket through the Live API, returning interim text as someone speaks and finalizing it after, with several options for detecting when speech stops.
Available in the Gemini API through Google AI Studio, Antigravity, and the Gemini Enterprise Agent Platform (public preview).
The same capability powers voice in the Gemini app on macOS and Gboard on Android, where you can dictate, summarize files, and rewrite copy into any window.