Mistral launches Voxtral 3 voice stack: speech models plus a real-time agent SDK
In short: Mistral AI released Voxtral 3, a set of speech recognition and speech synthesis models paired with an SDK that lets a single real-time pipeline listen, reason, call tools and speak back. The recognition models (Small, Mini and Realtime Transcribe 3) cover 50+ languages with speaker diarization, keyword-biasing for technical vocabulary, and transcription latency under 500ms, which Mistral claims beats Whisper, Scribe v2 and Gemini 3.5 Transcribe on its benchmarks. The speech synthesis model, Mini TTS 3, handles 35+ languages and 50+ dialects with emotion, pacing and emphasis that adapt to context, priced at $13 per million characters, and a separate Pro Voice feature can reproduce a person's voice and emotional range from just 30 minutes of recorded audio. All components can run through Mistral's EU-hosted API or be deployed on a customer's own servers.
This summary was generated automatically by AI from Mistral AI's publication. It is our own text, not a copy of the original — facts, figures and quotes belong to the source, linked above and below.
What changed?
- 1New transcription lineup: Voxtral Small Transcribe 3, Voxtral Mini Transcribe 3 (4B parameters, $0.0035/min), Voxtral Realtime Transcribe 3 (under 500ms latency, 35+ languages)
- 2Recognition now spans 50+ languages, 24 of them EU official languages, with diarization, keyword biasing and automatic end-of-turn detection
- 3Mini TTS 3 adds 35+ languages and 50+ dialects with context-driven emotion, pacing and emphasis, priced at $13 per 1M characters
- 4Pro Voice feature clones a speaker's voice character and emotional range using only 30 minutes of sample audio
- 5A new Voice Agent SDK merges listening, reasoning, tool calling and speaking into one real-time loop for phone and web agents
- 6Offered via EU-hosted API or on-premise deployment, letting organizations control where audio is processed and retained
| Parameter | Before | Now |
|---|---|---|
| Languages (STT) | Not specified | 50+ languages including 24 EU languages |
| STT latency (realtime) | Not specified | Sub-500ms in 35+ languages |
| Mini Transcribe price | Not specified | $0.0035/min, 4B parameters |
| TTS languages/dialects | Not specified | 35+ languages, 50+ dialects |
| TTS price | Not specified | $13 per 1M characters |
| Voice cloning | Not specified | Pro Voice adaptation from 30 minutes of audio |
| Tool calling | Not specified | New Voice Agent SDK combines listening, reasoning, tool calling and speaking in real time |
Why it matters
This gives teams building phone-based assistants a single, benchmarked package for speech recognition, speech generation and tool-calling logic instead of stitching together separate vendors, with published latency and pricing figures to compare against.
What it means for AI agents and contact centers
For anyone running voice agents or contact-center automation, this is worth comparing against current STT/TTS providers on Lithuanian-language accuracy, diarization quality, sub-500ms latency claims, and the quoted $0.0035/min and $13/1M character pricing before considering a switch or pilot.
🧪 Worth Testing
The combination of multilingual accuracy claims, sub-500ms real-time transcription, emotion-adaptive TTS and an integrated tool-calling SDK makes it a strong candidate to benchmark against existing voice-agent pipelines for latency, cost and language coverage.
Sources
- Mistral AIOfficialPrimary sourceOriginal article →„Voxtral 3“20 Oct 2026, 15:00
- Published by source
- 20 Oct 2026, 15:00
- Found by our system
- 20 Oct 2026, 19:16
- Summary generated
- 20 Oct 2026, 19:19
This article was written by AI from the original source. Facts, numbers and prices come from the source; missing values are marked “Not specified”. Legal notice, copyright and privacy