About Voxtral

Open Speech Models for the Multilingual Future

Voxtral is an open-source speech AI project that provides state-of-the-art transcription and spoken-language understanding capabilities. We're building the future of speech AI, one open model at a time.

Our Mission

To democratize speech AI technology by providing open-source, high-performance models that anyone can use, modify, and deploy. We believe that the future of AI should be transparent, accessible, and community-driven.

Our Values

What Drives Us

The principles that guide our development and shape our community.

Open & Accessible

We believe speech AI should be accessible to everyone. That's why Voxtral is open-source with Apache 2.0 licensing, enabling developers worldwide to build, modify, and deploy without restrictions.

Privacy First

Your data belongs to you. With self-hosting options and transparent processing, you maintain complete control over your audio data and transcriptions.

Innovation Driven

We push the boundaries of what's possible in speech AI, from 32K token context windows to function calling from voice, always staying at the cutting edge.

Community Focused

Built by developers, for developers. Our community-driven approach ensures Voxtral evolves to meet real-world needs and use cases.

By the Numbers

Performance Metrics

Real results from real benchmarks and production deployments.

95.2%
Accuracy Rate
on Common Voice dataset
15+
Languages
with auto-detection
32K
Token Context
40+ minutes of audio
$0.001
Per Minute
transparent pricing
Our Journey

Development Timeline

Key milestones in Voxtral's development and growth.

2024

Voxtral Launch

Released the first open-source speech AI model with 32K token context window and built-in Q&A capabilities.

2024

Multilingual Expansion

Added support for 15+ languages with automatic detection, making speech AI truly global.

2024

Function Calling

Pioneered voice-to-action capabilities, enabling direct API calls and workflow automation from speech.

2024

SOTA Performance

Achieved 95.2% accuracy on Common Voice, surpassing Whisper large-v3 and other leading models.

Open Source & Community

Voxtral is built by the community, for the community. Our Apache 2.0 license ensures that our work remains free and accessible to everyone. Join thousands of developers who are building the future of speech AI together.

Ready to Get Started?

Experience the power of open-source speech AI. Try our demo or explore the documentation.