GlobalBatch
Drop one real file in below and see what comes back — no account needed to watch it land. Everything this deployment accepts, and what Mistral AI publishes about it, is under the console.
Your audio, through Mistral AI voxtral-mini-latest, now
Need the words as you speak, not after the file? Compare the providers live — same models, streaming, billed by the second.
Click or drag audio file here
Supports MP3, M4A, WAV, OGG
Up to 30 minutes of audio; longer recordings accepted up to 4 hours.
Enables raw parameter access for Mistral Voxtral Mini Transcribe 2 (global). Disables universal options.
Provides a hint for the minimum and maximum number of expected speakers to improve diarization accuracy.
Boosts the recognition probability of specific words or phrases, such as proper nouns or domain-specific terms. Provide one phrase per line.
What this deployment is
What it accepts
| Option | On this deployment |
|---|---|
| Speaker diarizationLabels who spoke when, so a two-person recording comes back as a conversation. | Yes |
| Speaker count hintTakes a minimum and maximum number of speakers to steer the labelling. | No |
| Custom vocabularyTakes a list of terms — product names, jargon — to bias recognition toward. | Yes |
| Automatic punctuationReturns punctuated, cased text rather than an unbroken lexical stream. | No |
| Smart formattingWrites numbers, dates and currency as a reader expects them: twenty dollars as $20. | No |
| Profanity filterMasks profane words in the returned transcript. | No |
What this deployment accepts through our API. A no means the option is not offered on this model here — not that the vendor has never shipped it.
Languages
13 languages on this deployment, as the vendor publishes them. It will also identify the language itself when you do not name one.
The vendor’s own parameters
| Parameter | Default | What it does |
|---|---|---|
temperature | unset | Sampling temperature from 0 to 1. Leave blank for the model default. |
target_streaming_delay_ms | unset | Realtime only: how long the vendor waits for context before transcribing. Its own examples run 240ms for speed and 2400ms for accuracy. Leave blank for the model default. |
Passed through untouched, spelled exactly as Mistral AI spells them. They are on the advanced panel of the console above.
The fare, and what it is worth
We have not read a published fare for this deployment, so there is nothing here we would stand behind. The rest of the sheet is on the comparison page, and Mistral AI states this one on its own pages. What we charge is on our pricing page.
What Mistral AI says about it
Batch transcription with segment timestamps, speaker diarization, and context biasing across 13 languages, and realtime transcription over the WebSocket endpoint under a separate model id
Mistral Voxtral Mini Transcribe 2 (global) documentation · Every model and region we test · Compare all 11 on one file