Voxtral Mini Transcribe 3

Changes from previous version

Same-day smaller file model next to Voxtral Small Transcribe 3: 4B parameters at $0.0035 per minute, versus $0.004 per minute for Small.

Release Summary

The 4B file transcription model in the Voxtral 3 family, in public preview as voxtral-mini-transcribe-3-0. Price is $0.0035 per minute. The audio docs list a 65,536-token context shared by the audio and the vocabulary list, while the model card lists 32k. This entry uses 65,536 for both context and output because Mistral publishes one shared figure and no separate output cap. FLEURS 45 macro WER is 4.6% with the language specified and 4.5% inferred. Diarization cpWER is 18.6%. The FLEURS 35 chart uses this model as the offline baseline at 4.0% WER. It sits beside Voxtral Small Transcribe 3 and the live Mini Transcribe Realtime 3.

Timeline

Other Voxtral versions