Mistral releases Voxtral, its first open source AI audio model

In an era where artificial intelligence is revolutionizing communication with machines, French startup Mistral is making significant strides in the audio sector. This week, the company introduced Voxtral, its inaugural open-source audio model, setting its sights on competing with established proprietary systems. Voxtral is being marketed as a groundbreaking solution designed specifically for businesses, promising to bring practical speech intelligence into production environments. Mistral emphasizes that this model offers a viable alternative to existing solutions that either come at a high cost or lack functionality. The company claims that Voxtral is available at “less than half the price” of similar options, making it an attractive choice for businesses. Its capabilities include transcribing up to 30 minutes of audio, while its underlying Mistral Small 3.1 model can comprehend audio for up to 40 minutes. Users can interact with the content by asking questions, generating summaries, or executing voice commands for actions such as API calls. Voxtral is also designed with multilingual support, capable of transcribing and understanding various languages, including English, Spanish, French, Portuguese, Hindi, German, Dutch, and Italian. Mistral is releasing two versions of Voxtral: the larger Voxtral Small, equipped with 24 billion parameters for robust production applications, and the Voxtral Mini, which has 3 billion parameters tailored for local and edge deployments. Additionally, there is a cost-effective API variant, Voxtral Mini Transcribe, optimized for transcription tasks and positioned to outperform OpenAI Whisper at a lower cost. Users can access a trial of Voxtral by downloading the API from Hugging Face or experimenting with the models via Mistral’s chatbot, Le Chat. The integration of the API into applications is priced at just $0.001 per minute. This launch follows Mistral's recent announcement of Magistral, its first series of reasoning models that enhance reliability through step-by-step problem-solving. As one of Europe's leading AI firms, Mistral continues to champion open-source initiatives in the AI landscape and is reportedly in discussions to secure up to $1 billion in funding from investors, including Abu Dhabi’s MGX fund.