News Summary

French AI startup Mistral has launched Voxtral, its first open-source audio model, designed to provide a cost-effective alternative to closed corporate systems. Voxtral aims to deliver usable speech intelligence in production, supporting multiple languages such as English, Spanish, French, and others. The model is available in two variants: Voxtral Small, for production-scale deployments with 24 billion parameters, and Voxtral Mini, optimized for local and edge deployments with 3 billion parameters. Mistral claims Voxtral offers transcriptions at less than half the price of competitors, with capabilities such as summarizing audio content and executing real-time actions. This release follows Mistral’s recent announcement of its Magistral reasoning models, highlighting the company’s commitment to open-source AI solutions.