Mistral released Voxtral 3: a transcription model, a 4B mini, a realtime version under 500ms, a text to speech model, and a voice agent SDK built on LiveKit. On FLEURS across 45 languages Mistral reports a 4.1% word error rate against 16.0% for Whisper large-v3. Transcription starts at $0.0035 a minute and speech synthesis costs $13 per million characters. Voice cloning works from about thirty minutes of audio. Mistral has not stated a licence, so whether these are open weights is not yet knowable.