Voxtral Mini 4B Realtime 2602
Mistral AI · released Jan 21, 2026 · mistralai/Voxtral-Mini-4B-Realtime-2602
Speech to textTranscribes a recording into words
- Type
- Open weightsApache License 2.0
- Languages
- 13
- Size
- 4.4B
Context measured in tokens
Our take
Written Aug 2, 2026Voxtral Mini is a compact downloadable speech-to-text model from Mistral AI that turns audio into written words. It supports thirteen languages and processes audio extremely fast on benchmark hardware, though its accuracy varies sharply with audio quality.
Pick this for free local deployment with a permissive licence that allows commercial use and redistribution. Use it for batch transcription where speed matters, or for clean read-aloud English where it achieves its lowest error rate. Skip it if you need hosted inference, are transcribing meetings or heavily accented speech, or need verified accuracy in any language other than English.
The case for it
- Processes an hour of audio in about thirty-four seconds on benchmark hardware, at over one-hundred-times real-time speed.
- Apache 2.0 licence permits commercial use, fine-tuning and redistribution without restriction.
- Strong on clean read-aloud English, with a word error rate of about one in sixty-two — roughly four times better than its own overall average.
- Thirteen declared languages including major European and Asian languages.
The case against it
- Accuracy collapses on challenging audio: recorded meetings and heavily accented speech see error rates more than seven times worse than clean read-aloud.
- No hosted inference options available; you must self-host.
- Every accuracy figure we hold is English-only; no source measures the other twelve declared languages.
How good is it?
TranscriptionTurning speech into text2.5 of 5Open ASR WER · 52nd of 74
93.6%
Misses roughly one word in 16, averaged over nine English test sets.
105×58th of 62
an hour of audio in 34 seconds, on the board's own hardware. Your machine will differ.
13
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.
Which languages ↓ ↑
English · French · Spanish · German · Russian · Chinese · Japanese · Italian · Portuguese · Dutch · Arabic · Hindi · Korean
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. 18.5 GB spare means a 10% error in the size would not change the answer.
GeForce RTX 5090 · 32 GB
Room to spare. 26.5 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. 1.7 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 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
- mistralai/Voxtral-Mini-4B-Realtime-2602
- Architecture
- Dense
- Modality record
- audio->text
- Catalogue slug
- mistralai-voxtral-mini-4b-realtime-2602