Mistral Launches Voxtral 3 Speech Models and a Voice-Agent API From $0.04 a Minute
Mistral released Voxtral 3 on Oct. 20: speech-to-text models covering 50+ languages, Voxtral Speech 3 (Mini TTS 3) for 35+ languages and 50+ dialects, and a LiveKit-based voice-agent SDK. Listed prices include $0.0035/min for Mini Transcribe batch and $0.04/min for the agent API with Mistral Small 4. Benchmark wins are Mistral's own.
Contact-centre voice agents are a volume market; Mistral is competing on hosting choice and list price. On Tuesday, October 20, Mistral released Voxtral 3: Voxtral Small Transcribe 3 and a 4B Mini Transcribe 3 for speech-to-text in 50+ languages with diarization and context biasing; Voxtral Realtime Transcribe 3 at sub-500ms latency in over 35 languages; and Voxtral Speech 3 (Mini TTS 3) for expressive text-to-speech in 35+ languages and 50+ dialects.
A voice-agent SDK built on LiveKit chains listening, reasoning with a model such as Mistral Small 4, tool calling and speech into one streaming pipeline. Mistral says each turn completes in under a second and works over SIP telephony and WebRTC. Listed prices include $0.0035/min (Mini Transcribe batch), $0.008/min (Mini Transcribe realtime), $13 per 1M characters (Speech 3), and $0.04/min for the agent API with Mistral Small 4, or $0.08/min with any other reasoning model. It runs on Mistral's European, US and global endpoints or on customers' own infrastructure.
Mistral claims state-of-the-art transcription and diarization, ahead of Whisper, Scribe v2 and Gemini 3.5 Transcribe. Those comparisons, and most of its TTS charts, are Mistral's own evaluations — several TTS panels are internal evals judged by a Gemini model.
Limits
- Transcription and TTS benchmark wins are vendor-reported; The Decoder restates the launch without independent tests.
- Realtime language count is stated as "over 35" in the FLEURS section (another blurb says 40+); we use 35+.