Voxtral-grade transcription, without the API key
Mistral's Voxtral speech models are making headlines — but they're sold as a developer API billed per minute of audio. If you just have recordings you want to read, upload them here and get clean text free. No key, no sign-up.
Voxtral is Mistral AI's speech model family, and its current generation is strong: per Mistral's published benchmarks, Voxtral Small Transcribe 3 leads Whisper large-v3, ElevenLabs Scribe v2 and Gemini 3.5 Transcribe on multilingual word error rate, across 50+ languages with speaker diarization and context biasing. It ships as part of a voice-agent SDK aimed at developers building real-time voice products.
What it isn't is a transcription tool for everyone else. Using Voxtral means a Mistral API account and per-minute billing — $0.0035 per minute on the batch Mini model, more for realtime and the voice-agent stack. Great for apps that transcribe audio all day; heavy machinery for the recording on your phone you'd like to read once.
This page is the no-pipeline option: upload a voice memo, meeting recording or interview and get a clean, punctuated transcript in seconds — free, three times a day, in 90+ languages with automatic detection. If you're a developer evaluating Voxtral's API, the rates and trade-offs are below; if you just want the text, you can stop reading and press upload.
How to transcribe here
Drag in a voice memo, meeting recording or interview — m4a, mp3, wav, mp4, ogg, webm and flac all work. Nothing to install, no account to create.
The transcript comes back punctuated and paragraphed, with automatic language detection across 90+ languages — including the European and Asian languages Voxtral advertises.
Grab the text and paste it anywhere — notes, docs, subtitles. Searchable text beats scrubbing a waveform at 2x speed.
When this beats wiring up an API
A lecture, a call, a field interview — transcribing it once here is faster than opening a Mistral account and reading API docs.
Voxtral's headline is 50+ languages; Whisper covers 90+ with automatic detection. Either way, upload and read — no SDK required.
Testing Voxtral against other engines? Transcribe the same file here for a free second opinion before you commit per-minute spend.
Thoughts captured on the go, turned into text you can edit, copy and act on — the reason this site exists.
Questions, answered
What is Voxtral transcription?
Voxtral is Mistral AI's speech model family. The current generation (Voxtral 3, announced October 2026) includes Voxtral Small Transcribe 3 and Mini Transcribe 3 for speech-to-text, plus TTS and a realtime voice-agent stack — 50+ languages, speaker diarization, context biasing and sub-500ms realtime latency. It's sold through Mistral's API, aimed at developers.
How much does Voxtral transcription cost?
Per-minute billing, verified against Mistral's announcement page on Oct 23, 2026: Voxtral Mini Transcribe 3 (batch) is $0.0035/min (about $0.21 per hour of audio), Voxtral Small Transcribe 3 is $0.0040/min, and Voxtral Mini Transcribe Realtime 3 is $0.008/min. The voice-agent API runs $0.04–0.08/min including the reasoning model.
Is Voxtral better than Whisper?
Mistral's published benchmarks say yes: on the multilingual FLEURS test, Voxtral Small Transcribe 3 scores lower word error rate than Whisper large-v3, Scribe v2 and Gemini 3.5 Transcribe. Independent leaderboards are still catching up. This tool runs on OpenAI's Whisper models — proven, accurate across accents and noise — and it's free to try, so you can compare on your own audio.
Can I use Voxtral for free?
You can test it in Mistral's Studio playground. The API itself is pay-as-you-go per minute with no free tier documented as of Oct 23, 2026. This page gives you three free transcriptions every day — no API key, no account, no per-minute metering.
What languages does Voxtral support?
Mistral lists 50+ languages for batch transcription — including all 24 official EU languages and low-resource Asian and Indic languages — 35+ for realtime, and 35+ languages with 50+ dialects for its TTS. This tool transcribes in 90+ languages with automatic detection.
Do I need an API key to transcribe here?
No. Upload the file, get the transcript. Your audio is processed in memory to produce the text and never stored — no copy of your recordings is kept.