Cohere speech-to-text model
#13 in Transcription (STT)Cohere · since 26. März 2026 · 2× · last seen Oct 20, 2026
24
Momentum
Cohere Transcribe (cohere-transcribe-03-2026) is an open 2B-parameter automatic speech recognition model (audio-to-text) from Cohere, released on March 26, 2026 under the Apache 2.0 license. It supports 14 languages and can be self-hosted, accessed via the Cohere API, or run on Model Vault.
Momentum trend
23.07.21.10.
Features
| Real-Time Streaming | No streaming: transcribes completed audio files (35-second chunks for long files) |
| Latency | Throughput 525 minutes of audio per minute; real-time factor up to 3x faster than comparable dedicated ASR models (vendor claim) |
| License | Apache 2.0 |
| Platform | Self-hosted (Hugging Face download), Cohere API (Audio Transcriptions), Model Vault |
| Price | Open weights free (Apache 2.0); API free with rate limits (trial); Model Vault Medium $3.75/hr or $2,500/month |
| Release Date | March 26, 2026 |
| Languages | 14: English, German, French, Italian, Spanish, Portuguese, Greek, Dutch, Polish, Chinese, Japanese, Korean, Vietnamese, Arabic |