Canary Qwen 2.5b
NVIDIA · released Jun 26, 2025 · nvidia/canary-qwen-2.5b
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 Aug 3, 2026Canary Qwen is a tiny downloadable speech-to-text model from NVIDIA that turns English audio into written text. It is extremely fast and permissively licensed, but its accuracy widens dramatically on accented or conversational audio and no commercial hosts currently offer it.
Pick this for local or self-hosted transcription where permissive licensing matters, or for batch processing of clean read-aloud or financial-call audio where speed is critical. Skip it if you need accented-speech or meeting transcription, multilingual support, or a managed hosted service.
The case for it
- Extremely fast batch transcription: roughly one hour of audio in about four seconds on benchmark hardware.
- Strong on clean, structured audio with about one word in eighty wrong on read speech.
- Permissive Creative Commons Attribution 4.0 licence allows commercial use and redistribution with attribution.
- Small enough for edge deployment at 2.6 billion parameters.
The case against it
- Accuracy collapses on accented and conversational audio: error rate is more than six times higher on recorded meetings and more than eight times higher on accented speech compared with clean read speech.
- Only English is measured; multilingual claims are unverified in our data.
- No commercial hosting available; you must self-host.
How good is it?
TranscriptionTurning speech into text4 of 5Open ASR WER · 17th of 74
94.9%
Misses roughly one word in 20, averaged over nine English test sets.
861×26th of 62
an hour of audio in 4 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, not to the board.
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.
These tests check whether a model follows instructions — a precondition for all the work above, but not a measure of how well that work is done, which is why they get no rating.
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?
- Fits in memory
- weights load entirely on the card
- Spills to system RAM
- some weights offload; much slower
- Too large
- will not load even with offload
- est
- size is calculated; the verdict could change by 10%
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.
Memory use by level
Against a 24 GB card.
All 3 sizes here are calculated, not measured. We hold no measured file for this model, so each size comes from the parameter count and every verdict above inherits that uncertainty.
Check against your own machine → · All 71 devices, with every size →
When we formed this view
Dates behind this page
Prices last checked 6h ago
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
- nvidia/canary-qwen-2.5b
- Modality record
- audio->text
- Catalogue slug
- nvidia-canary-qwen-2-5b