Mistral AI Announces Voxtral 3 Voice Agent SDK
Mistral AI has released Voxtral 3, a voice agent SDK that combines listening, reasoning, tool calling, and speaking into a real-time conversational experience supporting over 50 languages. The suite includes three transcription variants and Voxtral Speech 3 TTS featuring emotional adaptability. This release provides developers with a robust tool to build real-time voice agents that can interact with business systems, retrieve information, and execute tasks over phone or web channels. It represents a significant step forward in multilingual speech recognition and workflow automation. Voxtral 3 offers sub-500ms real-time latency and is available via Mistral's EU-hosted APIs or for self-deployment, ensuring data sovereignty. However, some community members noted that it is a closed model, continuing the trend of proprietary solutions in text-to-speech and speech-to-text deployments.
## BACKGROUND
Voice agent SDKs integrate speech-to-text (ASR), large language models for reasoning, and text-to-speech (TTS) to enable natural spoken interactions with AI. Tool calling allows these agents to dynamically execute external functions or query databases, bridging the gap between conversational chat and actual software execution.