New Transcription Model: Cohere Transcribe

A new transcription model

Whisper Memos now supports Cohere Transcribe, a new speech recognition model released on March 26, 2026. It currently ranks #1 on HuggingFace's Open ASR Leaderboard with an average word error rate of 5.42%, outperforming both OpenAI's Whisper and ElevenLabs Scribe.

In our testing, the difference is especially noticeable for longer recordings and noisy environments. If you've ever had a memo come back with odd words or missing sentences, this model should help.

14 languages supported

Cohere Transcribe supports 14 languages: English, German, French, Italian, Spanish, Portuguese, Greek, Dutch, Polish, Arabic, Vietnamese, Chinese, Japanese, and Korean. If your language is on this list, the model will be used automatically. For other languages, the app falls back to one of the existing models.

How to switch

Open Settings → AI Model and select Cohere Transcribe. It's available on all plans, no upgrade needed. If you're using Autodetect for your language, the app will route your memos to Cohere automatically when it detects a supported language.

All available models

With this addition, Whisper Memos now offers four transcription models to choose from:

• Cohere Transcribe — the most accurate model available, supports 14 languages. Available on all plans.
• OpenAI Whisper Turbo — the fastest option with near-instant results. Great for most use cases.
• OpenAI Whisper — slightly more accurate than Turbo, slightly slower. A good fallback if Turbo gives you trouble.
• ElevenLabs Scribe — strong accuracy across a wider range of languages, including less common ones. Available on the Pro plan.