Voxtral Mini (latest)

Voxtral Mini sits inside Mistral's Voxtral speech-to-text family, which was refreshed as the Transcribe 2 generation announced on February 4, 2026. Within that family, the Mini variant is positioned as the batch-oriented transcription model, built for offline or asynchronous workloads rather than live streaming. Its companion sibling, Voxtral Realtime, is a separate model aimed at live applications and ships under the Apache 2.0 open-weights license; the Mini model itself is delivered through Mistral's commercial surface and is paired with a Mistral Studio audio playground that exposes diarization and timestamps for quick testing.

For practical fit, Voxtral Mini is shaped around high-quality, structured speech-to-text rather than open-ended conversation. It produces transcripts with speaker diarization, context biasing, and word-level timestamps across thirteen languages, and Mistral positions it as a best-in-class efficiency option that delivers the lowest word error rate at a comparatively low price point. The result is a model well suited to call-center analytics, meeting transcription, podcast indexing, and other pipeline tasks where reliable multi-speaker transcripts and timing metadata matter more than real-time interactivity.