Mistral has released Voxtral 3, its latest suite of speech models, adding support for 35 languages and introducing a software development kit for building voice agents. The update significantly expands the company's multilingual capabilities and offers new tools for developers.
TL;DR
- Mistral's Voxtral 3 now supports 35 languages and offers real-time voice agent tools.
- The update includes new transcription and speech generation models with advanced features.
- Developers gain more control over speech generation, including emotion tags and pronunciation specifications.
What happened
Mistral, a French AI company, launched Voxtral 3 on October 20, 2026. This update includes three transcription models and two speech generation models. The new models support a broader range of languages and offer advanced features for real-time transcription and speech generation.
The transcription models include Voxtral Mini Transcribe 3 for recorded audio, Voxtral Small Transcribe 3 for higher accuracy, and Voxtral Mini Transcribe Realtime 3 for live transcription. The speech generation models, Voxtral TTS 3 and Voxtral TTS Realtime 3, provide text-to-speech capabilities for both pre-recorded and live conversations.
Why it matters
For developers, the expanded language support and advanced features in Voxtral 3 offer new opportunities to build multilingual voice agents and applications. The Voice Agents SDK simplifies the integration of transcription, language models, and speech generation into a single streaming pipeline.
Startups and investors should note Mistral's focus on private deployment and data control, aligning with its sovereign AI strategy. This emphasis on customer control and infrastructure highlights Mistral's commitment to enterprise solutions and data security.
Key facts
- Voxtral 3 supports 50 languages for transcription and 35 languages for speech generation.
- The real-time transcription model covers 45 languages and dialects.
- New features include speaker labeling, detection of speaking turns, and emotion tags for speech generation.
- Developers can specify pronunciation for names and specialized terms.
- The Voice Agents SDK supports browser applications and telephone integration for enterprise customers.
- Models are available through European, US, and global API endpoints, with on-premises and private cloud deployment options.
Context
Mistral's Voxtral 3 builds on the company's previous releases, including Voxtral Transcribe 2 in February 2026 and Voxtral TTS in March 2026. The update reflects Mistral's ongoing commitment to expanding its multilingual capabilities and providing advanced tools for developers.
The emphasis on private deployment and data control aligns with Mistral's EUR 3bn Series D funding in September 2026. This strategy positions Mistral as a key player in the enterprise AI market, focusing on customer control and data security.
