Cohere Transcribe
Cohere Transcribe is a two-billion-parameter speech recognition model. It converts audio to text in 14 languages and is released under Apache 2.0, with evaluation API access and a production Model Vault route.
Capabilities and limits
The model works best with a specified single language and does not automatically detect the language. Its Conformer encoder processes audio features before a smaller text decoder generates the transcript. Supported languages include English, Arabic, Chinese, Japanese, Korean and several European languages.Cohere Transcribe documentation
When this model fits
Practical guidance. Use Transcribe when the deliverable is readable speech text, such as an archive transcript or input to a later summary. Include recordings with the microphones, accents and background noise your users have. If the product needs a synchronized player or named speakers, plan those features separately. Review names and numerical details against the recording before treating them as reliable extracted data.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | API, Downloadable |
| Model identifiers | cohere-transcribe-03-2026 |
| Maximum API file size | 25 MBCohere Transcribe documentation |
| Weight license | Apache 2.0Cohere Transcribe documentation |
| Output | Text without timestamps or speaker diarizationCohere Transcribe documentation |
Related models
A 111-billion-parameter text model for business workflows in 23 languages.
Launch date not confirmedAPI · Downloadable- Cohere Transcribe Arabic
Cohere · Transcribe
Two-billion-parameter Arabic speech recognizer with regional dialect and Arabic-English code-switching support.
Launch date not confirmedAPI · Downloadable