# Voxtral Mini Transcribe Realtime 3

mistral, released 2026-10-20

**Verdict: conditional.** Conditional. Voxtral Mini Transcribe Realtime 3 is Mistral's streaming speech-to-text model for agents at $0.008 per minute, with latency configurable from 80 ms to 2,400 ms and built-in end-of-turn detection, context biasing and speaker diarization in one WebSocket session. The deciding caveat is that the model page labels it Public Preview and we have no independent latency or accuracy test. Suits voice-agent builders already on the Mistral API; use Mini or Small Transcribe 3 for recorded audio. Last verified 2026-11-04.

## Pricing

- Input: $0.008 / min of audio
- Output: Billed per minute (single rate)
- Context: Per-minute billing; streaming; docs: Public Preview

## Strengths

- Voice agents that need streaming transcription; the model page describes it as built for agentic use-cases with built-in end-of-turn detection, context biasing and speaker diarization in a single streaming session over WebSocket
- Tuning latency against accuracy; the model page gives a configurable latency range of 80 ms to 2,400 ms, and the launch post card says sub-500ms latency
- Recorded-audio transcription on a budget; Voxtral Mini Transcribe 3 is $0.0035/min against $0.008/min here
- Self-hosting from open weights; the pages we read state no open weights or licence and show a Premier badge
- Anything that needs the state-of-the-art claim or latency confirmed outside Mistral; both are the model page's own wording and we have not tested them.
- Production dependence on a settled status; the model page badges it Public Preview

## Sources

- [Mistral docs: Voxtral Mini Transcribe Realtime 3 model page](https://docs.mistral.ai/models/voxtral-mini-transcribe-realtime-3-0)
- [Mistral: Voxtral 3](https://mistral.ai/news/voxtral-3/)

Canonical: https://neomanex.com/models/voxtral-mini-transcribe-realtime-3
