Mistral Releases Voxtral, Its First Open Source Ai Audio Model

As AI systems go much capable, reside is accelerated becoming nan default measurement we pass pinch machines. French AI startup Mistral has jumped into nan audio title pinch its first unfastened model, aiming to situation nan power of walled-off firm systems pinch open-weight alternatives.  

On Tuesday, Mistral announced nan merchandise of Voxtral, its first family of audio models aimed astatine businesses.

The institution is pitching Voxtral arsenic nan first unfastened exemplary that’s tin of deploying “truly usable reside intelligence successful production.”

In different words, nary longer will developers person to take betwixt a cheap, unfastened strategy that fumbles transcriptions and doesn’t really understand what’s being said, and 1 that functions well, but is closed, leaving developers pinch a higher measure and little power complete deployment. 

For businesses, that intends Voxtral offers an affordable replacement that nan institution claims is “less than half nan price” of comparable solutions.

Image Credits:Mistral

Mistral says Voxtral tin transcribe up to 30 minutes of audio. Due to its LLM backbone, Mistral Small 3.1, it tin understand up to 40 minutes, allowing users to inquire questions astir nan audio content, make summaries, aliases move sound commands into real-time actions for illustration calling APIs aliases moving functions. Voxtral is besides multilingual, pinch nan expertise to transcribe and understand languages including English, Spanish, French, Portuguese, Hindi, German, Dutch, and Italian.

The institution is offering up 2 variants of its “speech knowing models”. The first, Voxtral Small, has 24B parameters for production-scale deployments, and is competitory pinch ElevenLabs Scribe, GPT-4o-mini, and Gemini 2.5 Flash. 

The second, Voxtral Mini, has 3 cardinal parameters for section and separator deployments. There’s besides an ultra-cheap, stripped-down, accelerated API type of nan 3B exemplary called Voxtral Mini Transcribe that is optimized for transcription-only usage cases and promises to outperform OpenAI Whisper for little than half nan price.

Users tin effort Voxtral for free by downloading nan API connected Hugging Face aliases testing nan models successful Mistral’s chatbot Le Chat. Integrating nan API into applications starts astatine $0.001 per minute, according to nan company. 

The motorboat comes a period aft Mistral announced Magistral, its first family of reasoning models that activity done problems step-by-step for improved reliability. 

Mistral, 1 of nan apical AI firms successful Europe, is well-known for its defense pushing unfastened root AI models. Earlier this month, TechCrunch reported that nan institution is successful talks to raise up to $1 cardinal successful equity from investors for illustration Abu Dhabi’s MGX fund.