Mistral AI Voxtral Release

Mistral AI has introduced Voxtral, a family of state-of-the-art speech understanding models designed to bridge the gap between low-accuracy open-source ASR systems and expensive proprietary APIs. Voxtral provides high-accuracy transcription and native semantic understanding, available in two sizes—a 24B variant for production-scale applications and a 3B variant for local and edge deployments—both released under the Apache 2.0 license.

Key Capabilities and Technical Specifications

Voxtral models integrate speech processing with the text understanding capabilities of the Mistral Small 3.1 and Ministral backbones. This allows the models to perform complex tasks without needing to chain separate Automatic Speech Recognition (ASR) and language models.

Audio Processing and Context

  • Context Window: Voxtral supports a 32k token context length, enabling the transcription of audio files up to 30 minutes and the understanding of audio content up to 40 minutes.
  • Native Multilingualism: The models feature automatic language detection and high performance across widely used languages, including English, Spanish, French, Portuguese, Hindi, German, Dutch, and Italian.

Functional Capabilities

  • Integrated Q&A and Summarization: Users can ask questions directly about audio content or generate structured summaries natively.
  • Direct Function Calling: Voxtral can trigger backend functions, workflows, or API calls based on spoken user intents, removing the need for intermediate parsing steps.
  • Text Proficiency: Because they retain the capabilities of their language model backbones, Voxtral models can serve as drop-in replacements for Mistral Small 3.1 and Ministral.

Performance Benchmarks

Voxtral demonstrates superior performance across transcription and audio understanding tasks compared to both open-source and proprietary alternatives.

Speech Transcription

Voxtral outperforms Whisper large-v3, GPT-4o mini Transcribe, and Gemini 2.5 Flash across all evaluated tasks. It achieves state-of-the-art results on English short-form transcription and the Mozilla Common Voice 15.1 benchmark. In the FLEURS benchmark, Voxtral Small outperforms Whisper on every task and reaches state-of-the-art performance in several European languages.

Audio Understanding and Translation

Voxtral Small is competitive with GPT-4o-mini and Gemini 2.5 Flash in audio understanding tasks. It also achieves state-of-the-art performance in Speech Translation on the FLEURS-Translation benchmark.

Deployment and Pricing

Voxtral is available through multiple channels to accommodate different deployment needs:

  • Local Deployment: Both the 24B and 3B models are available for download via Hugging Face.
  • API Access: A transcribe-optimized version, Voxtral Mini Transcribe, is available via the Mistral AI API starting at $0.001 per minute. Mistral AI claims this version outperforms OpenAI Whisper at less than half the cost.
  • Le Chat: Voxtral is being integrated into Le Chat's voice mode for web and mobile users.

Enterprise and Future Roadmap

For enterprise users, Mistral AI offers private production-scale deployments, domain-specific fine-tuning (e.g., for legal or medical contexts), and dedicated integration support.

Future Feature Updates

Mistral AI is working on expanding Voxtral's audio capabilities to include:

  • Speaker segmentation
  • Word-level timestamps
  • Non-speech audio recognition
  • Audio markups for emotion and age

Research Documentation

Detailed technical information regarding the research and development of Voxtral is available in the official research paper (arXiv:2507.13264).

Sources

Related

  • Dispatch
    Mistral AI Voxtral Transcribe 2 Release

    Mistral AI has released Voxtral Transcribe 2, featuring Voxtral Mini Transcribe V2 for high-efficiency batch processing and Voxtral Realtime for ultra-low latency live transcription.

  • Dispatch
    Mistral AI Voxtral TTS Release

    Mistral AI has released Voxtral TTS, a 4B-parameter multilingual text-to-speech model that provides high-naturalness voice generation and low-latency streaming for enterprise voice agents.

  • Dispatch
    Mistral Small 3.1 Release Notes

    Mistral AI has released Mistral Small 3.1, an Apache 2.0 licensed model featuring multimodal understanding, a 128k token context window, and performance that surpasses GPT-4o Mini and Gemma 3.

  • Dispatch
    Codestral Release Notes

    Mistral AI has released Codestral, a 22B open-weight generative AI model specifically designed for code generation across 80+ programming languages.

  • Dispatch
    Pixtral Large Release Notes

    Mistral AI has released Pixtral Large, a 124B open-weights multimodal model that achieves state-of-the-art performance on MathVista, DocVQA, and VQAv2 while maintaining the text capabilities of Mistral Large 2.