French startup Mistral has announced an open-source audio model that can not only transcribe speech but also analyze its content through integration with a large language model.
French startup Mistral has announced an open-source audio model that can not only transcribe speech but also analyze its content through integration with a large language model.
As TechCrunch reports, the Voxtral model is designed for business integrations and supports working with audio up to 30 minutes long. Thanks to the connection with the LLM Mistral Small 3.1, it can not only recognize speech, but also analyze it: for example, answer questions about the content of what was heard, generate short sentences or carry out voice instructions, such as calling an API or launching functions in the application.
The model is multilingual: Voxtral supports English, Spanish, French, Portuguese, Italian, German, Dutch, and Hindi (unfortunately, Ukrainian is not available). This allows you to build localized voice interfaces and use it in different regions.
Mistral offers three Voxtral options:
The API is already available on Hugging Face, and testing is also possible in the Le Chat chat. The model can be embedded in web services, applications, or corporate systems.
The launch of Voxtral is part of Mistral’s strategy to develop open solutions in the field of AI. In June, the company introduced the Magistral reasoning model, which solves problems through step-by-step logical analysis. And in July, TechCrunch reported that the startup is in talks to attract up to $1 billion in investments, including from Abu Dhabi-based MGX Fund.
Mistral is positioning itself as an open alternative to players like OpenAI, Google, and Meta, which develop closed models with limited access to developers. Voxtral is the company’s first attempt to influence the rapidly growing segment of voice AI, which has until now remained under the control of large corporations.
We previously reported how Mistral launched Europe’s first artificial intelligence model that uses logical thinking to generate answers, as the French startup tries to keep up with American and Chinese competitors.
https://ykuntsevych.medium.com/europes-data-vision-dataspaces-for-zero-trust-ai-infrastructure-92947ddb9602