On October 21, 2026, Mistral AI officially released the brand-new Voxtral3 voice model series and Voice Agent SDK. This product matrix covers the Voxtral Small Transcribe3 transcription model, the Voxtral Mini TTS3.0 speech synthesis model, and a LiveKit-based Voice Agent SDK. They are available via API or can be self-hosted on private clouds and the user's own infrastructure, aiming to provide high-performance real-time streaming voice interaction capabilities for call centers, help desks, virtual assistants, and hands-free driving systems.
In terms of technical performance, Voxtral Small Transcribe3 supports more than 50 languages and offers voice segmentation, contextual biasing, and end-of-turn detection. In the FLEURS chart, its forced error rate and inference error rate are 4.1% and 4.0%, on par with Microsoft AI Transcribe2, and outperforming Gemini3.5Transcribe, Scribe v2, GPT Transcribe, and Whisper large-v3.
Among them, the Voxtral Mini Transcribe3 with 4 billion parameters costs $0.0035 per minute for batch usage, while the real-time version costs $0.008 per minute with latency below 500 milliseconds. Meanwhile, Voxtral Mini TTS3.0 offers speech synthesis in more than 35 languages and over 50 dialects, priced at $13 per million characters, supporting tone adaptation, prompt instructions, and Pro Voice cross-language voice cloning based on 30 minutes of audio.
The SDK built on LiveKit integrates transcription, reasoning, tool calling, and voice generation into the same real-time streaming pipeline, enabling general-purpose models to reason and call tools autonomously while handling conversations, keeping conversation turns under one second, and supporting automatic agent interruption when callers interject. Analysts believe that Mistral's release balances model performance with enterprise-grade deployment control, and through its modular architecture and private deployment support, further accelerates the large-scale adoption of voice AI agents in complex business scenarios.