Voxtral Mini is the small sibling of Mistral's audio family, published alongside Voxtral Small in July 2025 for local and edge deployment. At three billion parameters it runs on a single consumer GPU or a well-specified laptop, which matters wherever recordings cannot leave the building — hospitals, law firms, public administration — or where a robot has to hear and react without a network round trip. It keeps the family's defining trait: the same model that writes down what was said can also answer questions about it. Mistral additionally offers a transcription-only endpoint built on this model at the lowest price point in its API.