Mistral AI
Open and portable generative AI for devs and businesses
5.0•42 reviews•4.3K followers
Open and portable generative AI for devs and businesses
5.0•42 reviews•4.3K followers
- We’re committed to empower the AI community with open technology. Our open models sets the bar for efficiency, and are available for free, with fully permissive license.
- Our optimized commercial models are designed for performance and are available via our flexible deployment options.
This is the 25th launch from Mistral AI. View more
Mistral Voxtral 3
Launching today
Voxtral 3 is Mistral’s new family of speech-to-text and text-to-speech models, plus a voice agent SDK built on LiveKit. Transcription covers 50+ languages with diarization and context biasing, and realtime transcription runs under 500ms. TTS adapts emotion and pacing, with voice cloning from 30+ minutes of audio. Batch transcription starts at $0.0035 per minute, and the models are available through Mistral’s API.
Free Options
Launch tags:Artificial Intelligence
Launch Team
Dropbox SignSimple, secure eSignatures for the way your team works.
Promoted
ChatGPT by OpenAI
Claude by Anthropic
Perplexity for Mac
Voxtral 3 is Mistral’s new family of speech models: speech-to-text, text-to-speech and a voice agent SDK, all through Mistral’s API.
Problem: Building a voice agent usually means gluing together separate pieces for transcription, reasoning and speech. Transcription also tends to break down on real calls, with several speakers, domain terms and many languages, and robotic voices make conversations feel stiff.
Solution: Voxtral 3 covers the full loop. There are batch and realtime transcription models, a text-to-speech model, and an SDK built on LiveKit that combines listening, reasoning, tool calling and speaking in one real-time pipeline.
What makes it different: Each stage of the voice agent is a separate model, so you can swap the LLM, the voice or the language. Transcription handles context biasing with keyterms or free-form text, no fine-tuning needed, and the TTS adapts emotion, pacing and emphasis to the conversation. Per Mistral, realtime transcription runs under 500 ms, and a full agent turn completes in under a second.
Key features:
Voxtral Small Transcribe 3 for the highest accuracy, with speaker diarization and context biasing
Voxtral Mini Transcribe 3 for batch transcription (4B parameters)
Voxtral Mini Transcribe Realtime 3 for agents, with sub-500 ms latency
Voxtral Mini TTS 3 with emotion, pacing and emphasis controls, and voice cloning from about 30 minutes of audio
Transcription in 50+ languages, and TTS in 35+ languages and 50+ dialects
Voice agent SDK with function calling, web search, SIP telephony and WebRTC
Self-deployment and on-prem through Mistral sales
Pricing per Mistral: batch transcription from $0.0035 per minute, realtime transcription at $0.008 per minute, TTS at $13 per 1M characters, and the voice agent API from $0.04 per minute.
Who it’s for: Developers and teams building voice agents, call transcription, or multilingual voice apps who want one provider for the whole pipeline.
French guy here so obviously biased, but happy to see Mistral push on speech. Does it cope with people switching between French and English mid sentence? Because that's every call I'm on.
Can I now use more languages than only English and French with the API?