Mistral releases a new open-source model for speech generation 26/03/2026 às 11:30 Mistral’s new speech model can run on a smartwatch or a smartphone.