Mistral AI has entered the audio AI market with the release of Voxtral, an open-source family of large audio models built for transcription, summarization, and audio-based Q&A. Released under the permissive Apache 2.0 license, Voxtral is production-ready and targets enterprises building speech interfaces without vendor lock-in. The lineup includes Voxtral Small for server deployment and Voxtral Mini designed for lightweight, edge-device inference.

Performance benchmarks position Voxtral as a strong competitor to proprietary services. According to Mistral, Voxtral Small outperforms OpenAI’s Whisper large-v3, GPT-4o mini Transcribe, Gemini 2.5 Flash, and ElevenLabs Scribe. The model supports transcription of up to 30 minutes and audio understanding across 40-minute contexts. It offers multilingual capabilities in English, French, Spanish, Portuguese, Hindi, and more. Mistral claims Voxtral’s API costs less than $0.001 per minute, slashing costs by more than half compared to commercial rivals.

The launch positions Mistral as a serious contender in the speech AI market. Voxtral provides an open-source option for businesses building voice applications, from automated call center transcription to real-time assistants and media analysis. By combining high accuracy, long context windows, and cost efficiency, Voxtral challenges incumbents relying on proprietary models and creates fresh opportunities for AI builders.