← All models

Voxtral Small Transcribe 3

Audio Preview Full comparison ↗

mistral/voxtral-small-transcribe-3 · by Mistral AI · speech-transcription-model

Pricing — 1 offering(s)

Mistral AI

About this offering: Mistral's flagship speech-to-text model, in Public Preview since 2026-10-20, at $0.004 per minute of audio. It transcribes 50 languages with speaker diarization, timestamps and context biasing, and also does translation, speech Q&A and summarization, per Mistral. Mistral also lists a cached-input rate of $0.0004 per minute, which isn't tracked as a separate price here.

Audio minutes

  • $0.004 / minute Current 2026-10-20 → present

Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.

See how Voxtral Small Transcribe 3 fits into a cost-aware routing setup

See how →

Capability profile

How Voxtral Small Transcribe 3 rates across core capability dimensions, with the task-level evidence behind each rating.

transcription accuracy Strong

Mistral reports a 4.1% macro-average WER on 45 FLEURS languages (language specified), tied with Microsoft AI Transcribe 2 at the top of its chart, ahead of Gemini 3.5 Transcribe, Scribe v2, GPT Transcribe and Whisper large-v3 (16.0%). Vendor-reported.

language support Strong

50 languages, including the 24 official EU languages, per Mistral's launch post; vendor-reported.

speaker diarisation Strong

Native diarization; Mistral reports 18.4% cpWER, the lowest in its chart. Vendor-reported.

real time streaming Weak

Offline (file) transcription; Mistral's realtime model is the separate Voxtral Mini Transcribe Realtime 3.

custom vocabulary Strong

Context biasing; Mistral reports 94.6% keyphrase recall on Earnings-22 with context (81.4% without). Vendor-reported.

Benchmarks

Benchmark Score Config Source
FLEURS 45-language WER (language specified) 4.1 % Mistral's launch post: macro-average WER (CER for JA/KO/TH/ZH) over the 45 FLEURS languages all seven compared models support; 4.0% with the language auto-detected. Vendor-reported. source ↗
Diarization cpWER 18.4 % Mistral's launch post: macro-average cpWER over AMI-IHM, AMI-SDM, Switchboard, CallHome and AISHELL-4. Vendor-reported. source ↗
Earnings-22 keyphrase recall (with context) 94.6 % Mistral's launch post; 81.4% without context biasing. Vendor-reported. source ↗

Operator guidance

Choose Voxtral Small Transcribe 3 when accuracy and diarization matter most in batch transcription: it's at the top of Mistral's own FLEURS and diarization charts, at about the price of Nova-3 pre-recorded. Choose the Mini model for lower cost. Its figures are vendor-reported and it's a Public Preview.

Use cases

Limitations

Citations