Voxtral 3: Voice Agent SDK and Speech Models in 50 Languages (mistral.ai)

Voxtral 3 has launched a sophisticated voice agent SDK along with advanced speech models capable of real-time, multi-step conversations in over 50 languages. This rollout integrates accurate transcription, natural speech synthesis that adapts emotionally, and seamless tool integration for workflow automation, making it a significant advancement for applications in contact centers, virtual assistants, and hands-free copilots. With support for various languages, including low-resource Asian and Indic dialects, the SDK combines listening, reasoning, and action execution into a single, low-latency pipeline, addressing the historical challenges of real-time voice interactions. Technical highlights include the Voxtral Small Transcribe 3, achieving state-of-the-art multilingual transcription with capabilities such as diarization and context biasing, outpacing existing models like Whisper and Gemini. Additionally, the Voxtral Mini TTS 3 enhances speech synthesis with expressive, context-aware responses across multiple languages and dialects. With options for EU-hosted API or self-deployment, organizations gain control over sensitive data while benefiting from a system that can accurately interpret nuanced intent and execute multi-step actions. This marks a leap forward in creating more reliable and engaging voice interactions, positioned for diverse deployment scenarios.