New Transcription Model: Gemini 3.5 Transcribe

A new transcription model

Whisper Memos now supports Gemini 3.5 Transcribe, Google's latest speech recognition model. It detects the spoken language on its own, handles recordings up to one hour, and is available on every plan — no upgrade needed.

It's a strong all-rounder: accurate on everyday recordings, comfortable with mixed languages, and fast enough that you rarely wait for a transcript.

Better meeting transcripts

Gemini also knows who is speaking. If you pick it as your model, meetings up to 30 minutes go to Gemini and come back split by speaker.

Meetings you don't pick a model for keep using ElevenLabs Scribe, which is still our most dependable option for separating speakers. The same goes for meetings over 30 minutes, and for meetings where you use custom words — Gemini can't combine a custom vocabulary with speaker separation yet.

How to switch

Open Settings → AI Model and pick Gemini 3.5 Transcribe. Every new recording will use it; existing memos stay as they are. Recordings longer than an hour fall back to the usual models automatically.

All available models

• Automatic — picks the best model for each recording based on language, length, and audio quality.
• OpenAI Whisper — balanced accuracy and speed, a good default for most recordings.
• Cohere Transcribe — the most accurate option for its 14 supported languages.
• Grok Speech to Text — xAI's model, solid across multiple languages.
• Gemini 3.5 Transcribe — Google's latest, with automatic language detection and speaker separation in Meeting mode.
• ElevenLabs Scribe — the broadest language coverage, including less common languages. Available on Pro.

Update Whisper Memos to try Gemini 3.5 Transcribe on your next recording.