With the rapid development of artificial intelligence technology, speech is quickly becoming the main way we communicate with machines. French startup Mistral has officially released its first open-source audio model - Voxtral, aiming to break the monopoly of large enterprises' closed systems and provide developers with a more flexible and cost-effective alternative.

Mistral claims that Voxtral is the first open-source model that can provide "truly usable speech intelligence" in real-world applications. This means developers no longer have to make difficult choices between low-cost open-source systems and efficient but closed solutions. With the advantage of "less than half the price," Voxtral offers companies a more economical option.

image.png

According to Mistral, Voxtral can transcribe audio up to 30 minutes long. Due to its basis on the large language model Mistral Small3.1, users can understand audio content as long as 40 minutes. Users not only can ask questions related to the audio content, but also generate summaries, and even convert voice commands into real-time operations, such as calling an API or performing specific functions. In addition, Voxtral supports multiple languages, including English, Spanish, French, Portuguese, Hindi, German, Dutch, and Italian.

Mistral provides two variants of "speech understanding models." The first is Voxtral Small, with 24 billion parameters, suitable for production-level deployment, and competing with ElevenLabs Scribe, GPT-4o-mini, and Gemini2.5Flash. The second is Voxtral Mini, with 3 billion parameters, suitable for local and edge deployment. There is also an ultra-economic version with 300 million parameters called Voxtral Mini Transcribe, optimized for transcription scenarios, promising performance that surpasses OpenAI's Whisper, while costing less than half of it.

Users can download the API of Voxtral for free through Hugging Face, or test it in Mistral's chatbot Le Chat. According to the company, the integration cost of the API starts from $0.001 per minute. This release comes just a month after Mistral launched its reasoning model Magistral, and these models improve reliability by solving problems step by step.

As one of Europe's leading artificial intelligence companies, Mistral has always been actively promoting the development of open-source AI models. It is worth mentioning that recently there were reports that Mistral is in talks with investors to raise up to $1 billion in funding, including the MGX fund from Abu Dhabi.