Mistral Launches Voxtral: Affordable Open Source AI Audio Models for Businesses

As AI systems continue to advance, speech is rapidly becoming the primary method of communication between humans and machines. French AI startup Mistral has entered the audio landscape with its first open-source model, aiming to challenge the dominance of proprietary corporate systems by offering more accessible and flexible alternatives. On Tuesday, Mistral announced the release of Voxtral, its inaugural family of audio models designed for businesses. The company positions Voxtral as the first open-source model capable of delivering “truly usable speech intelligence in production.” This means developers no longer need to choose between a low-cost, open system that struggles with accuracy and understanding, and a more expensive, closed system that performs well but leaves them with limited control and higher costs. Voxtral offers a cost-effective solution, with the company claiming it is “less than half the price” of similar closed-source options. It supports various functionalities, including transcribing up to 30 minutes of audio. Thanks to its large language model (LLM) backbone, Mistral Small 3.1, it can handle up to 40 minutes, enabling users to ask questions, generate summaries, and execute real-time actions such as calling APIs or running functions. Voxtral is also multilingual, capable of transcribing and understanding languages such as English, Spanish, French, Portuguese, Hindi, German, Dutch, and Italian. Mistral is releasing two variants of its “speech understanding models.” The first, Voxtral Small, features 24 billion parameters and is designed for production-scale deployments. It competes with models like ElevenLabs Scribe, GPT-4o-mini, and Gemini 2.5 Flash. The second variant, Voxtral Mini, has 3 billion parameters and is suitable for local and edge deployments. Additionally, there is a cost-optimized version called Voxtral Mini Transcribe, which is streamlined for transcription-only use cases and promises better performance than OpenAI’s Whisper for a fraction of the price. Developers can test Voxtral for free by downloading the API on Hugging Face or using it in Mistral’s chatbot, Le Chat. For commercial applications, integrating the API into software starts at just $0.001 per minute of audio. This launch follows the release of Magistral, Mistral’s first family of reasoning models, which was announced a month ago. Magistral is designed to solve complex problems step-by-step, enhancing reliability and performance. Mistral is one of Europe’s leading AI firms and has gained recognition for its strong advocacy of open-source AI models. Earlier this month, TechCrunch reported that the company is in talks to raise up to $1 billion in equity from investors, including Abu Dhabi’s MGX fund. By providing affordable, high-performance, and open-source options, Mistral aims to democratize access to advanced AI technologies and empower businesses and developers to innovate more freely.