Voxtral 3: Mistral's voice agent stack — speech models plus SDK in 50+ languages
Mistral's Voxtral 3 family pairs three speech-to-text models and a text-to-speech model with a LiveKit-based voice agent SDK, streaming listening, reasoning, tool calls and spoken replies as one pipeline for $0.04 a minute.
Why it matters
A full voice-agent stack from a European lab at commodity prices: transcription from $0.0035 a minute, expressive TTS with voice cloning, and an SDK that turns the whole loop into one streaming call.
What you could build with it
An indie developer could build a multilingual voice receptionist for small clinics: patients book and reschedule by phone in their own language, with the agent checking a real calendar through tool calls, and measure how many calls complete without a human callback.
Does it hold up?
Usable in public preview on the Mistral API since 20 October, with a $0.04/min voice-agent tier; independent coverage flags unresolved limits — phone calls are enterprise-only and inbound-only, and agents cannot hand off to a human — so production phone use is still gated.
Built with Voxtral 3
- Mistral Voxtral 3 Launches With $0.04/Min Voice Agentsarticle · Breakdown of the model lineup and published prices, plus the caveats: conflicting language counts in Mistral's docs and vendor-reported latency claims.
- Voxtral 3 voice agents: what businesses should testarticle · Business evaluation of the stack, the public-preview status, and what remains unproven before production use.
- What Is Voxtral 3? Mistral Speech Models and Voice Agentsarticle · Full explainer of the five models, the LiveKit-based SDK, the $0.04/min voice-agent API, and the limits worth knowing before building on it.
Learn more
First spotted on article: source.
More AI models
VoiceStudio: open-source, fully-local ElevenLabs alternative for voice cloning, dubbing, and audiobooksdebpalash/VoiceStudio is a desktop app that runs voice cloning, voice design, video dubbing, dictation, transcription…models · JEV 0.75Edge0-145B: open 145B MoE that runs on a single consumer GPUEdge0 open-sourced 145B and 120B latent-MoE models derived from Qwen3.8-Flash-Next, plus an inference framework that…models · JEV 0.74Strata: run a 125B model on a gaming PCFree, MIT-licensed local inference engine that runs Qwen3.8-Flash-Next (125B MoE, about 6B active per token) on…models · JEV 0.73Google AI Edge Gallery: offline speech recognition and synthesis on-deviceGoogle added fully offline speech recognition and text-to-speech to AI Edge Gallery, with ML Kit APIs on Android and…models · JEV 0.73Soup: fine-tune an 8B LLM on a 4 GB laptop GPU from one YAML fileAn open-source CLI that fine-tunes LLMs from a single YAML file, with layer streaming that trains an 8B model on a 4…models · JEV 0.73Tencent Hunyuan Hy3: open 295B MoE with 21B active under Apache 2.0Tencent open-sourced Hy3, a 295B-parameter mixture-of-experts model with only 21B parameters active per token, under…models · JEV 0.73
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.