Voxtral 3 Release - FLEURS 45 - Inferred: leaderboard
Voxtral 3 release evaluation (2026-10-20): exact chart marks retain model, series and score. FLEURS 45 macro-averages WER, using CER for Japanese, Korean, Thai and Chinese; language forced/inferred runs stay separate. Diarization averages cpWER over AMI-IHM, AMI-SDM, Switchboard, CallHome and AISHELL-4 with 30-minute clips. Context-bias recall uses Earnings22. Realtime latency configurations remain distinct. TTS tests retain their release-specific populations and judge; chart x-axis prices are never scores. WER/CER (%).
Metric: WER/CER (%). Source: mistral.ai. Saturation forecast: Rough model projection: around 2026. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Voxtral Small Transcribe 3 | 4.1 | |
| 2 | Microsoft AI Transcribe 2 | 4.1 | |
| 3 | Voxtral Mini Transcribe 3 | 4.6 | |
| 4 | Scribe v2 | 4.9 | |
| 5 | Gemini 3.5 Transcribe | 6.1 | |
| 6 | GPT Transcribe | 6.3 | |
| 7 | Whisper large-v3 | 16 |
Interactive version: theaggregate.ai/benchmark?slug=voxtral-3-release-fleurs-45-inferred · How It Works · Data refreshed daily, snapshot 2026-10-21.