Cohere Transcribe 03 2026
Cohere Labs · released Mar 24, 2026 · CohereLabs/cohere-transcribe-03-2026
Speech to textTranscribes a recording into words
- Type
- Open weightsApache License 2.0
- Languages
- 14
- Size
- 2.1B
Context measured in tokens
Our take
Written Sep 17, 2026Cohere Transcribe 03 2026 is a downloadable English speech-to-text model that gets 4.7% of words wrong across nine English test sets, better than most models on that measure. The licence allows commercial use, changes and redistribution, and there is no host to fall back on, so running it yourself is the route.
Use it to transcribe English recordings where accuracy matters and you can run the model yourself, or to work through a large backlog of audio. Skip it if you need speaker labelling, timestamps or streaming, if you need a language other than English, or if you want a hosted offer rather than running it yourself.
The case for it
- Among the more accurate English speech-to-text models we list: 4.7% of words wrong averaged over nine English test sets, against a field where the best is 3.6% and the middle 5.2%.
- Accurate on clean read-aloud audio at 1% of words wrong, better than most models on that condition, where the best is 0.9% and the middle 1.5%.
- Holds up on harder recordings: 7.9% of words wrong on podcasts and video, 7.9% on accented speech and 7% on meetings, each better than most models on that condition.
- Fast enough to make a backlog practical: 907 times real time on the leaderboard's own hardware, so an hour of audio takes about 4 seconds there, though your machine may differ.
The case against it
- Every accuracy figure is English only: 14 languages are listed, but no source we hold measures any language but English, so accuracy in the others is unverified in our data.
- The boards measure words recognised, not who said them, so speaker labelling, timestamps and streaming are all unmeasured here.
- We list no host for it, so running it yourself is the only route we can point you to.
How good is it?
An open transcription model for turning meeting recordings and clear read-aloud audio into written text.
- transcribing recordings of meetings in a roomRecorded meetings · 7th of 92
- transcribing clear recordings of people reading aloudClean read speech · 6th of 92
TranscriptionTurning speech into text4 of 5Open ASR WER · 22nd of 76
95.3%
Misses roughly one word in 21, averaged over nine English test sets.
907×32nd of 74
an hour of audio in 4 seconds, on the board's own hardware. Your machine will differ.
14
Listed on the model card. The accuracy above is English only.
Percentage of words wrong on each set, lower better. Bars are scaled to this model's own worst case; the placing beneath each rate is against every model measured on that set.
Which languages ↓ ↑
Arabic · German · Greek · English · Spanish · French · Italian · Japanese · Korean · Dutch · Polish · Portuguese · Vietnamese · Chinese
The figures above come from the Open ASR Leaderboard, an independent public test that runs every model on the same recordings. It is the only measurement of transcription quality we know of, so there are no other scores to show.
Each of these is the same transcription job on a different kind of recording, so together they say where it holds up and where it slips — not how closely it follows an instruction.
Every published score for this model9 scoresEvery figure we hold, from 9 boards, with who ran it and a link to the source — including the boards no rating above is built on.
Can you run it yourself?
Comfortable fit
GeForce RTX 4090 · 24 GB
Room to spare. 20.2 GB spare means a 10% error in the size would not change the answer.
GeForce RTX 5090 · 32 GB
Room to spare. 28.2 GB spare means a 10% error in the size would not change the answer.
Apple M2 (8-core GPU, 8GB unified) · 8 GB
Room to spare. 3.4 GB spare means a 10% error in the size would not change the answer.
These cards answer whether Cohere Transcribe 03 2026 loads, not how fast it transcribes. Throughput figures for a transcription model come from its text decoder, so treat this as a fit answer rather than a speed one.
Memory use by level
Against a 24 GB card.
What is quantisation? →This model on every device we track71 devicesThe Q4 build most people download, on each device: what the weights come to, how much context the memory leaves, and whether it runs. Smallest device that runs it first. This is a fit answer: whether it loads, not how fast it transcribes.
When we formed this view
Recent changes
What moved
first indexed by our pipelineEach date is the day we first saw the change, or the day the maker announced it.
What we do not know about this model yet
- We hold no measured file for it, so all 3 sizes on this page are calculated from the parameter count.
- Nothing we hold says whether an endpoint streams, so we do not show it either way.
Licence and identifiers
What the licence allowsApache License 2.0, what it allows commercially, and the identifiers you need to pull this model — its Hugging Face repo, our slug and a machine-readable card.
Licence
Apache License 2.0
Fully permissive: commercial use, redistribution, and derivatives allowed. Requires attribution and a copy of the license. Includes an express patent grant.
Identifiers
- Hugging Face
- CohereLabs/cohere-transcribe-03-2026
- Takes in, gives back
- Audio in, text out
- Catalogue slug
- coherelabs-cohere-transcribe-03-2026