STT 2.6b en
Kyutai · released Jun 6, 2025 · kyutai/stt-2.6b-en
Speech to textTranscribes a recording into words
- Type
- Open weightsCreative Commons Attribution 4.0
- Languages
- 1
- Size
- 2.6B
Context measured in tokens
Our take
Written Sep 17, 2026STT 2.6b en is a downloadable English speech-to-text model from Kyutai, and speed is its selling point: on the leaderboard's own hardware it works through an hour of audio in about 27 seconds. Accuracy is middling overall, and it is weakest on accented speech and meeting-room recordings, so it suits clean, clearly recorded English rather than messy real-world conversations.
Reach for it when you are transcribing clean, clearly recorded English — read-aloud material, podcasts, video — and want to clear a long backlog, or when you would rather run the model on your own machine. Its licence allows commercial use, changes and redistribution. Skip it if you need accurate transcription of accented speakers or meeting-room audio, or if you need any language other than English.
The case for it
- Fast enough to work through a large audio backlog: 133 times real time on the leaderboard's own hardware, an hour of audio in about 27 seconds, so a long queue of recordings becomes practical to process.
- Accurate on clean read-aloud audio, at 1.4% of words wrong, better than most models on this condition.
- Solid on podcast and video audio, at 8.2% of words wrong, better than most models on this condition.
- The licence allows commercial use, changes and redistribution (Creative Commons Attribution 4.0).
The case against it
- Struggles with accented speakers, at 8.9% of words wrong, worse than most models on this condition.
- Weak on meeting-room recordings, at 10.5% of words wrong, worse than most models on this condition.
- English only: one language, and every accuracy figure we hold is English.
How good is it?
TranscriptionTurning speech into text3 of 5Open ASR WER · 49th of 76
94.4%
Misses roughly one word in 18, averaged over nine English test sets.
133×66th of 74
an hour of audio in 27 seconds, on the board's own hardware. Your machine will differ.
1
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.
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. 19.9 GB spare means a 10% error in the size would not change the answer.
GeForce RTX 5090 · 32 GB
Room to spare. 27.9 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. 3.1 GB spare means a 10% error in the size would not change the answer.
These cards answer whether STT 2.6b en 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 allowsCreative Commons Attribution 4.0, 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
Creative Commons Attribution 4.0
Permissive content license: any use with attribution. Common for datasets and some model weights.
Identifiers
- Hugging Face
- kyutai/stt-2.6b-en
- Architecture
- Dense
- Takes in, gives back
- Audio in, text out
- Catalogue slug
- kyutai-stt-2-6b-en