Voxtral 3 Voxtral 3 is Mistral AI's family of state-of-the-art Transcribe and Speech models that power a new voice agents API. It includes multilingual transcription with diarization and context biasing, expressive TTS across 35+ languages, and a LiveKit-based SDK that streams listen-reason-speak with tool calling in a single sub-second pipeline, available via hosted API or self-hosted.
Visit productFeatured media
Loading media…
View on XCosigns Received
“Can an STT model make sense of a 135 year old recording? I tested @MistralAI new Voxtral transcription model on a noisy French phonograph recording of Gustave Eiffel from 1891. It matched 100 of 107 words in INA’s published excerpts (6.5% WER), while transcribing 75 additional words omitted from the reference. It handled Eiffel’s dictation well but struggled with proper names and some guest dialogue. A fascinating test of modern speech recognition on audio from 1891.”
“Mistral 这次的 Voxtral 3，我觉得最有用的不是转写，是 TTS 能逐句标情绪。拿它给我的 App 推广视频配了段旁白，[像不像真人 / 哪里露馅]。要是真能用，做 UGC 视频又能省一笔配音钱。”
“Quick test of Voxtral 3 on release night. The end-of-turn detection built into the STT is really fast! - No more cut-offs mid-hesitation - Dictating a phone number goes through in one turn - 0.6 s less silence before each reply, on an STT/LLM/TTS stack that started at 2.3 s - Bonus: the voxtral-mini-tts-3-0 TTS no longer makes up sentences on short words, with the same response time Small sample, but very promising.”
“Extremely cool! I was using the voice API of Mistral all the time; much better in French than Claude. It just got updated with Voxtral 3 🚀”