Voxtral transcribes at the speed of sound

simonwillison.net2026年02月04日 22:42

Voxtral transcribes at the speed of sound

Mistral just released Voxtral Transcribe 2 - a family of two new models, one open weights, for transcribing audio to text. This is the latest in their Whisper-like model family, and a sequel to the original Voxtral which they released in July 2025.

Voxtral Realtime - official name Voxtral-Mini-4B-Realtime-2602 - is the open weights (Apache-2.0) model, available as a 8.87GB download from Hugging Face.

You can try it out in this live demo - don't be put off by the "No microphone found" message, clicking "Record" should have your browser request permission and then start the demo working. I was very impressed by the demo - I talked quickly and used jargon like Django and WebAssembly and it correctly transcribed my text within moments of me uttering each sound.

The closed weight model is called voxtral-mini-latest and can be accessed via the Mistral API, using calls that look something like this:

curl -X POST "https://api.mistral.ai/v1/audio/transcriptions" \
  -H "Authorization: Bearer $MISTRAL_API_KEY" \
  -F model="voxtral-mini-latest" \
  -F file=@"Pelican talk at the library.m4a" \
  -F diarize=true \
  -F context_bias="Datasette" \
  -F timestamp_granularities="segment"

It's priced at $0.003/minute, which is $0.18/hour.

The Mistral API console now has a speech-to-text playground for exercising the new model and it is excellent. You can upload an audio file and promptly get a diarized transcript in a pleasant interface, with options to download the result in text, SRT or JSON format.

Screenshot of a speech-to-text transcription interface for a file named "Pelican talk at the library.m4a". The toolbar shows "Speech to text" with Code, Transcribe, and Download buttons. The transcript shows timestamped segments from 5:53 to 6:53 with a speaker icon, reading: "5:53 – 6:01 So pelicans love to, they're very good at getting the most they can out of the topography when they're flying. 6:01 – 6:06 And our winds come in from the northwest and they hit those bluffs and they're deflected up. 6:07 – 6:18 And they will sit right, they'll fly north into a wind like five feet off those bluffs, but just five or ten feet off the surface because the winds dissipate. 6:19 – 6:22 And they will surf that bluff all the way north. 6:23 – 6:30 So you'll see a wind from the north at 15 miles an hour, and the pelicans are flying north into that wind and not flapping their wings. 6:31 – 6:33 And it's one of the coolest things. 6:33 – 6:35 You can only find it on San Francisco Coast. 6:36 – 6:39 Where right where the bluffs are steep. 6:41 – 6:43 Pacifica, you can find them there. 6:43 – 6:51 They like their, what we call pier bums, which are typically pelicans that have, are in some sort of trouble. 6:51 – 6:53 They're unable to catch food." The segment at 6:41–6:43 is highlighted in yellow. An audio waveform is shown at the bottom with a playhead near 6:40. Stats in the lower right show 53.90s, 7946.00s, and #45833.

Via Hacker News

Tags: ai, generative-ai, llms, hugging-face, mistral, speech-to-text