All tasks / Models
cohere-transcribe-03-2026
Cohere's Apache-2.0 open-weights speech recognition model, also served through Cohere's API.
- Cohere
- Open weights
- Released Mar 2026
Where it ranks
- BTranscriptionModels · Local · 4.6% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)
Facts
- Vendor
- Cohere
- Family
- Cohere Transcribe
- Released
- Mar 2026
- Weights
- Open: download and run it yourself
- License
- Apache-2.0
- Parameters
- 2.07B
- GPU memory to run locally
- ~6 GB (Derived: 2.07B parameters (Hugging Face safetensors count) x 2 bytes = ~4.1 GB of bf16 weights.)
- Access
- Free (open weights)
- Last verified
- 18 Oct 2026
Benchmark results
| Benchmark | Result | Entry on the leaderboard | As of | Source |
|---|---|---|---|---|
| Artificial Analysis Speech to Text: AA-WER v2 (non-streaming) | 4.6% | transcribe-03-2026AA-WER v2 (non-streaming), provider Cohere; embedded value 0.0457 converted to percent and rounded; page shows no last-updated date. | 18 Oct 2026 | artificialanalysis.ai |
| Hugging Face Open ASR Leaderboard (English, short-form average WER) | 5.8% | CohereLabs/cohere-transcribe-03-2026Average WER over the board's default datasets (English short-form, version 17-10-2026); rank 11; read from the Space's initial table data. | 17 Oct 2026 | huggingface.co |
Where you can use it
- –MuesliMuesli (open source)