Model

Voxtral Mini Transcribe V2

Family
Voxtral Transcribe 2
Context window
Open weights
No
Release date
2026-02-04

Benchmark notes

Batch transcription model supporting 13 languages with speaker diarization, context biasing (up to 100 terms) and word-level timestamps; reported ~4% word error rate on FLEURS benchmark; priced at $0.003/min.