Hi everyone!
Mistral AI’s new Voxtral models are a big step for open-source speech AI. They’re built to go beyond simple transcription and into true understanding.
This means the AI can answer questions about audio, summarize conversations, and even trigger functions directly from voice commands.
It’s great that they’ve released both a powerful 24B model and a smaller 3B version for on-device use under an Apache 2.0 license. This makes high-quality speech understanding much more accessible. For access, you can run it locally, use their API, or try it in Le Chat’s voice mode which is rolling out in the coming weeks.