Mistral releases Voxtral, its first open source AI audio model

As the AI systems are able to be more able, the speech quickly becomes the default way of communicating with the machines. French AI Startup Mistral Open-Wait options have jumped into the audio race with its first open model to challenge the domination of Wallid-off corporate systems.

On Tuesday, the first family of the audio model announced the publication of Vxtral, targeting the Mysterial business.

The company is the first open model that is pitching Vxtral that is capable of deploying “truly useful speech intelligence in production”.

In other words, developers no longer have to choose in a cheap, open system that wash the transcripts and does not really understand what is being said and it works well, but it is closed, with a higher bill and low control over the deployment.

For business, it means that the vostral offers an affordable option that the company claims that the comparative solutions are “less than half the price”.

Mistral says Vxtral can replicate 30 minutes audio. Due to its LLM spine, Mistral Small 3.1It can be understood by up to 40 minutes, to ask users to ask questions about audio content, to create summaries or to make voice commands to call API or turn into real-time verbs like running functions. Voxtral is also multilingual with the ability to replicate and understand languages including English, Spanish, French, Portuguese, Hindi, German, Dutch and Italian.

The company is providing two variants of its “speech understanding models”. First, the Voxtral smaller has 24B parameters for the production-scale deployment and it is competitive with the Eggs script, GPT-4-Mini and Jemi 2.5 flash.

Second, there are 3 billion parameters for the deployment of Voxtral mini, local and edge. Also, the 3B model called Voxtral Mini Transkrib has a superpower, stripped-down, rapid API version, which is favorable for transcript-cable use and promises the opening whipper to be less than half the price.

Users can try to hug the faces or try to download API by checking models in Mistral Chattbott Le Chat. According to the agency, APIs are integrated into applications, starting at $ 0.001 per minute.

The launch comes a month later Mistral Magistral announcedIts first family of rational models that work through step by step for better reliability.

The Mistral Open Source, one of the top AI companies in Europe, is well known to him to push the AI models. Earlier this month, TechCrunch reported that the company was in talks to raise 1 billion dollars in Equity From investors like Abu Dhabi MGX Fund.