Voxtral
Mistral's speech models: open weights to self-host, or a per-minute API
Quick Overview
| Company | Mistral (simplified joint stock company, trading as Mistral AI) |
|---|---|
| Category | Speech-to-Text APIs |
| Headquarters | Paris, France |
| Founded | Not stated |
| EU Presence | Yes - France (EU) |
| Data Location | not stated on the pages we read (the regional inference page was not read) |
| Open Source | Apache 2.0 for the original Voxtral 24B and 3B weights |
| GDPR Compliant | Yes |
| Self-Hosting | Yes - the original Voxtral weights can be downloaded from Hugging Face and run locally or on-premises |
| Main Features | File and live transcription, speaker labels, timestamps, custom vocabulary, automatic language detection, audio question answering, text-to-speech |
| Pricing | API: Voxtral Mini Transcribe 3 $0.0035 per minute, Voxtral Small Transcribe 3 $0.004 per minute, Voxtral Mini Transcribe Realtime 3 $0.008 per minute (Mistral pricing page, USD or EUR toggle) |
| Best For | Developers who want a cheap per-minute API now and the option to run open weights themselves later |
| Replaces | Deepgram, AssemblyAI |
Detailed Review
Voxtral is a family of speech models published by Mistral, the company behind the Mistral AI platform. The legal notice names Mistral as a simplified joint stock company registered in Paris under number 952 418 325, with offices at 15 rue des Halles.
Mistral released the first Voxtral models in July 2025 in two sizes, 24B and 3B, under the Apache 2.0 licence. The current audio documentation covers Voxtral 3, which transcribes recorded files and live streams and sits next to text-to-speech and voice agents on the same API.
What do Voxtral models transcribe, and how?
The Mistral audio documentation says Voxtral 3 models transcribe recorded files and live streams, with speaker labels, timestamps and custom vocabulary. Audio can also be sent to a chat model to summarize a recording or ask questions about it.
The 2025 launch post lists automatic language detection and strong results in English, Spanish, French, Portuguese, Hindi, German, Dutch and Italian. It states a 32k token context that handles audio up to 30 minutes for transcription.
How much does the Voxtral API cost?
The Mistral API pricing page lists Voxtral Mini Transcribe 3 at $0.0035 per minute and Voxtral Small Transcribe 3 at $0.004 per minute. The realtime model, Voxtral Mini Transcribe Realtime 3, is listed at $0.008 per minute. The page offers a USD or EUR toggle.
Cached input is priced at one tenth of these rates. The launch post claims Voxtral Mini Transcribe beats OpenAI Whisper at less than half the price, which is Mistral's own benchmark claim.
Voxtral as a European alternative to Deepgram
Voxtral gives teams a French contracting entity and open weights they can run in their own infrastructure, which Deepgram's hosted API does not offer in the pages we read. Deepgram is stronger on a dedicated contact-centre focus and a single platform with text-to-speech and voice agents.
Mistral also offers text-to-speech through Voxtral TTS, so a voice agent can be built on one European vendor. Compare with Voxist for EUR billing and a free tier.
Limitations to Consider
The pages we read do not state where API audio is processed, and the Mistral legal notice shows the website itself is hosted by Netlify in San Francisco. The 30 minute transcription limit comes from the 2025 launch post and may differ for Voxtral 3. Voxtral is one product line inside a large AI company, so support and speech-specific tooling are less specialised than at a pure speech vendor.
Who Should Use Voxtral
Developers and product teams who already use or consider the Mistral platform and want inexpensive transcription with an exit route to self-hosted open weights.
Pros and Cons
Pros
- Per-minute rates published, starting at $0.0035
- Open Apache 2.0 weights for the original models
- Live streams, speaker labels and custom vocabulary documented
- French company registered in Paris
- Transcription, text-to-speech and chat on one API
Cons
- API data location not stated on the pages we read
- Benchmark claims come from Mistral itself
- Prices are shown in USD by default
- No contact-centre specific tooling mentioned
Alternatives to Voxtral
Other European speech-to-text apis worth comparing before you decide:
See all European speech-to-text apis or compare them directly against Deepgram on our alternatives page.
Frequently Asked Questions
Yes. Voxtral is made by Mistral, a company registered in Paris, and it offers transcription through an API as well as downloadable open weights. Deepgram is a California company.
The Mistral pricing page lists Voxtral Mini Transcribe 3 at $0.0035 per minute, Voxtral Small Transcribe 3 at $0.004 per minute and the realtime model at $0.008 per minute.
The original Voxtral 24B and 3B models are released under Apache 2.0 and can be downloaded from Hugging Face. Mistral describes running them locally or on-premises.