Mistral AIAudio matchupSpeech to text
Voxtral Mini Transcribe 3vs.Voxtral Small Transcribe 3
Which model is better?
Voxtral Small Transcribe 3 does better in independent tests.
Pick Voxtral Mini Transcribe 3 for
Pick Voxtral Small Transcribe 3 for
Shared tests
Independent evaluators and leaderboards only, the newest shared snapshot per evaluator. With 95 % ranges a lead counts when they do not overlap; for hit rates, when it holds in a 95 % test. Without any uncertainty figure at most “slightly ahead”.
Artificial Analysis
Voxtral Mini Transcribe 3 slightly aheadVoxtral Mini Transcribe 3 is clearly ahead in 0 of 14 subtests and slightly in 12, Voxtral Small Transcribe 3 in 0 clearly and slightly in 2.
Artificial Analysis batch speed factor: p05
as of October 24, 2026
Artificial Analysis daily batch median speed factor: 2026-10-21
as of October 24, 2026
Artificial Analysis batch speed factor: q25
as of October 24, 2026
Show all 14 subtests
| Subtest | Voxtral Mini Transcribe 3 | Voxtral Small Transcribe 3 | Verdict |
|---|---|---|---|
| Artificial Analysis batch speed factor: p05 | 67.17 | 48.41 | Voxtral Mini Transcribe 3 slightly ahead |
| Artificial Analysis daily batch median speed factor: 2026-10-21 | 76.98 | 58.38 | Voxtral Mini Transcribe 3 slightly ahead |
| Artificial Analysis batch speed factor: q25 | 71.01 | 56.22 | Voxtral Mini Transcribe 3 slightly ahead |
| Artificial Analysis daily batch median speed factor: 2026-10-24 | 81.91 | 65.55 | Voxtral Mini Transcribe 3 slightly ahead |
| Artificial Analysis Speech-to-Text WER: AgentTalk (lower is better) | 2.0 % | 1.6 % | Voxtral Small Transcribe 3 slightly ahead |
| Artificial Analysis batch speed factor: p95 | 83.81 | 71.67 | Voxtral Mini Transcribe 3 slightly ahead |
| Artificial Analysis daily batch median speed factor: 2026-10-22 | 71.47 | 61.41 | Voxtral Mini Transcribe 3 slightly ahead |
| Artificial Analysis batch speed factor: median | 74.75 | 64.47 | Voxtral Mini Transcribe 3 slightly ahead |
| Artificial Analysis batch speed factor: q75 | 76.98 | 66.70 | Voxtral Mini Transcribe 3 slightly ahead |
| Artificial Analysis daily batch median speed factor: 2026-10-23 | 72.88 | 63.42 | Voxtral Mini Transcribe 3 slightly ahead |
| Artificial Analysis daily batch median speed factor: 2026-10-20 | 74.71 | 66.62 | Voxtral Mini Transcribe 3 slightly ahead |
| Artificial Analysis Speech-to-Text AA-WER index (lower is better) | 2.2 % | 2.1 % | Voxtral Small Transcribe 3 slightly ahead |
| Artificial Analysis Speech-to-Text WER: Earnings22-Cleaned-AA (lower is better) | 3.4 % | 3.5 % | Voxtral Mini Transcribe 3 slightly ahead |
| Artificial Analysis Speech-to-Text WER: VoxPopuli-Cleaned-AA (lower is better) | 1.4 % | 1.5 % | Voxtral Mini Transcribe 3 slightly ahead |
Source · latest October 24, 2026
Open Danish ASR Leaderboard
Voxtral Small Transcribe 3 aheadVoxtral Small Transcribe 3 is clearly ahead in 9 of 13 subtests and slightly in 1, Voxtral Mini Transcribe 3 in 1 clearly and slightly in 1. 1 is level.
Open Danish ASR Leaderboard FLEURS Danish da_dk CER
95 % ranges 1.9 % to 2.5 % and 1.4 % to 1.8 %, 930 tasks, lower is better, as of October 24, 2026
Open Danish ASR Leaderboard FLEURS Danish da_dk WER
95 % ranges 5.4 % to 6.4 % and 4.0 % to 4.9 %, 930 tasks, lower is better, as of October 24, 2026
Open Danish ASR Leaderboard CoRal v3 read-aloud WER
95 % ranges 20.8 % to 21.7 % and 17.2 % to 18.1 %, 9,122 tasks, lower is better, as of October 24, 2026
Show all 13 subtests
| Subtest | Voxtral Mini Transcribe 3 | Voxtral Small Transcribe 3 | Verdict |
|---|---|---|---|
| Open Danish ASR Leaderboard FLEURS Danish da_dk CER (lower is better) | 2.2 % | 1.6 % | Voxtral Small Transcribe 3 ahead |
| Open Danish ASR Leaderboard FLEURS Danish da_dk WER (lower is better) | 5.9 % | 4.4 % | Voxtral Small Transcribe 3 ahead |
| Open Danish ASR Leaderboard CoRal v3 read-aloud WER (lower is better) | 21.3 % | 17.6 % | Voxtral Small Transcribe 3 ahead |
| Open Danish ASR Leaderboard CoRal v3 read-aloud CER (lower is better) | 8.4 % | 7.3 % | Voxtral Small Transcribe 3 ahead |
| Open Danish ASR Leaderboard Common Voice Danish WER (lower is better) | 10.5 % | 9.0 % | Voxtral Small Transcribe 3 ahead |
| Open Danish ASR Leaderboard FTSpeech test_balanced CER (lower is better) | 6.8 % | 7.6 % | Voxtral Mini Transcribe 3 ahead |
| Open Danish ASR Leaderboard macro mean WER (lower is better) | 17.0 % | 15.4 % | Voxtral Small Transcribe 3 ahead |
| Open Danish ASR Leaderboard macro mean CER (lower is better) | 8.7 % | 8.2 % | Voxtral Small Transcribe 3 ahead |
| Open Danish ASR Leaderboard CoRal v3 conversation WER (lower is better) | 35.1 % | 33.2 % | Voxtral Small Transcribe 3 ahead |
| Open Danish ASR Leaderboard CoRal v3 conversation CER (lower is better) | 22.4 % | 21.2 % | Voxtral Small Transcribe 3 ahead |
| Open Danish ASR Leaderboard Common Voice Danish CER (lower is better) | 3.8 % | 3.5 % | Voxtral Small Transcribe 3 slightly ahead |
| Open Danish ASR Leaderboard FTSpeech test_balanced WER (lower is better) | 12.2 % | 12.7 % | Voxtral Mini Transcribe 3 slightly ahead |
| Open Danish ASR Leaderboard inference speed | 5.80 | 5.70 | level |
Source · latest October 24, 2026
Vendor figures
Figures from vendor reports that cover both models. They do not count toward the verdict.
Show 4 vendor figures
| Figure | Source | Voxtral Mini Transcribe 3 | Voxtral Small Transcribe 3 |
|---|---|---|---|
| Artificial Analysis weighted WER index (lower is better) | Mistral Voxtral 3 announcement and benchmark tables | 2.2 % | 2.1 % |
| Diarization benchmark cpWER (lower is better) | Mistral Voxtral 3 announcement and benchmark tables | 18.6 % | 18.4 % |
| FLEURS 45 forced transcription WER (lower is better) | Mistral Voxtral 3 announcement and benchmark tables | 4.5 % | 4.0 % |
| FLEURS 45 inferred transcription WER (lower is better) | Mistral Voxtral 3 announcement and benchmark tables | 4.6 % | 4.1 % |
All data side by side
Prices are verbatim from the provider, because audio models bill per minute, character or token.
| Voxtral Mini Transcribe 3 | Voxtral Small Transcribe 3 | |
|---|---|---|
| Task | Speech to text | Speech to text |
| Developer | Mistral AI | Mistral AI |
| Status | Public preview | Public preview |
| Released | October 20, 2026 | October 20, 2026 |
| Runs locally | no | no |
| Parameters | 4 B | n/a |
| Price | USD 0.0035 per audio minute; cached rate USD 0.00035 per minute | USD 0.004 per audio minute; cached rate USD 0.0004 per minute |
| Popularity (out of 100) | 0 | 0 |