Mistral launches Voxtral 3 speech models and a voice-agent SDK that can answer phone calls
Announced · Published · By Contractor AI Shop
Mistral released Voxtral 3: speech-to-text models covering 50+ languages, the Voxtral Mini TTS 3 text-to-speech model (35+ languages, 50+ dialects) and a LiveKit-based voice-agent SDK that streams listening, reasoning, tool calls and speech in one pipeline. It runs through Mistral's API or self-deployed.
Key facts
- Mistral says the agent answers and places calls over SIP, runs in browsers and apps over WebRTC, and completes each turn in under one second; the latency figure is Mistral's. Mistral AI
- Mistral lists transcription at $0.0035/min (Mini, batch), $0.004/min (Small) and $0.008/min (Realtime), text-to-speech at $13 per 1M characters, and its voice agent API from $0.04/min with Mistral Small 4. Mistral AI
- Mistral says Voxtral 3 can be self-deployed on your own infrastructure; for the full stack on-premises or in a private cloud it asks you to contact its team. Mistral AI
- Mistral's model docs list Voxtral Mini Transcribe Realtime 3 as a public preview, released Oct 20, 2026. Mistral AI (docs)
- The Decoder reports agents can tap external systems during live conversations, for example to book appointments or check invoices. The Decoder (Matthias Bastian)
- TAO Media notes context biasing lets developers supply key terms to improve recognition of specialized vocabulary without fine-tuning. TAO Media (Bart and Tristan Hillerich)
Our take: what it means for contractors
This is the building kit for the "AI answers the shop phone and books the estimate visit" setup. Self-deployment matters if you don't want customer calls in someone else's cloud. Mistral's own demo is medical appointment booking, not trades, so it needs a hands-on test before we recommend it.
What we don't know yet
- Real-world latency on phone lines (the under-one-second turn figure is Mistral's)
- Hardware needed to self-host
- How well it handles jobsite noise and trade vocabulary
- When the realtime model leaves public preview
Sources
- Mistral AI · Voxtral 3: Voice Agent SDK & Speech Models in 50+ Languages · 2026-10-20 · Primary
- Mistral AI (docs) · Voxtral Mini Transcribe Realtime 3 · 2026-10-20 · Primary
- The Decoder (Matthias Bastian) · Mistral AI launches Voxtral 3, a full voice agent stack that listens, thinks, and talks in real time · 2026-10-20 · Secondary
- TAO Media (Bart and Tristan Hillerich) · Mistral Releases Voxtral 3 With Multilingual Speech Models and LiveKit Voice Agent SDK · 2026-10-21 · Secondary
- Business+IT (SB Creative, Japanese) · ミストラルAI、音声エージェントSDKと「Voxtral 3」発表 50言語超の文字起こしに対応 · 2026-10-21 · Secondary