Mistral releases Voxtral 3 speech and voice-agent models

Mistral has released Voxtral 3 for transcription, speech synthesis, and real-time voice agents via EU APIs or self-hosting.

Mistral released Voxtral 3, a family of speech-to-text and text-to-speech models plus a voice-agent SDK that streams listening, reasoning, tool calling, and speech. Voxtral Small Transcribe 3 and a 4-billion-parameter Mini variant cover more than 50 languages, including 24 official EU languages, with diarization, context biasing, and end-of-turn detection. Voxtral Mini TTS 3 covers more than 35 languages and 50 dialects; Pro Voice adaptation is described as using 30 minutes of audio, and text-to-speech includes emotional adaptation. The Decoder says the SDK is based on LiveKit and can call external tools during a call. Deployment is through Mistral’s EU-hosted API or customer infrastructure; Mini Transcribe 3 is priced at $0.0035 per minute. Mistral says Small beats Whisper, Scribe v2, and Gemini 3.5 Transcribe on a 45-language FLEURS set.