Voxtral 3

Accurate voice agents now handle complex, multi-step tasks in dozens of languages, directly within your conversations.

Mistral AI has released Voxtral 3, a new voice agent SDK and a suite of speech models supporting over 50 languages. This technology aims to enable natural, real-time conversational experiences for applications like contact center agents, virtual assistants, and hands-free copilots. Voxtral 3 integrates accurate speech-to-text (STT) and adaptive text-to-speech (TTS) with robust tool integration for multi-step workflow automation. Key features include state-of-the-art multilingual transcription with diarization and context biasing, natural-sounding speech synthesis that adapts to emotional context in over 35 languages, and a voice agent SDK that combines listening, reasoning, and action execution within a single streaming pipeline. The models are available via Mistral's EU-hosted API or for self-deployment, emphasizing data sovereignty and low-latency processing. Performance benchmarks indicate Voxtral 3 models outperform competitors like Whisper and Gemini 3.5 Transcribe in areas such as multilingual transcription accuracy and diarization.

The release of Voxtral 3 signifies a significant advancement in the practical application of voice AI, moving beyond simple command-and-control to sophisticated conversational agents capable of complex reasoning and tool utilization. By offering low-latency, real-time streaming pipelines that integrate transcription, reasoning, and synthesis, Mistral AI is lowering the barrier to entry for building highly functional voice-powered applications. This enhanced capability directly impacts how businesses can automate customer interactions, streamline internal workflows, and develop new forms of human-computer interaction, particularly in multilingual environments where data sovereignty is a concern.

This development positions Mistral AI as a strong contender in the enterprise voice AI market, challenging established players by offering a comprehensive, sovereign-ready solution. The emphasis on adaptability, both in speech emotion and contextual understanding, suggests a future where voice agents are not just functional but also more engaging and human-like. Organizations seeking to deploy voice AI for sensitive applications or those operating under strict data residency requirements will find the self-deployment option particularly attractive, enabling greater control and customization.