Mistral Launches Voxtral: Its First AI Audio Model

Mistral has released Voxtral, its first audio AI model family, under the Apache 2.0 license.

The release includes three models: Voxtral Small (~24B parameters) for high-performance use, Voxtral Mini (~3B) for on-device applications, and Voxtral Mini Transcribe, designed for efficient transcription.

The models support up to 32,000 tokens of audio input and are capable of tasks beyond transcription, including summarization, question answering, and function calling.

Voxtral supports multiple languages and is available both for local download and via API.

According to Mistral, it outperforms Whisper large-v3 and other proprietary models on key benchmarks.

You may also want to check out some of our other recent updates.

Subscribe to Vavoza Insider to access the latest business and marketing insights, news, and trends daily. 🗞️