Mistral releases Pixtral 12B its first multimodal model

French AI startup Mistral has made a significant leap with the release of Pixtral 12B, a model that processes both images and text.

This 12-billion-parameter model is now available for download and fine-tuning on platforms like GitHub and Hugging Face.

Mistral’s Groundbreaking Release

French AI startup Mistral has launched its first model that processes both images and text. Named Pixtral 12B, this new model is approximately 24GB in size. It’s available for download and fine-tuning on GitHub and Hugging Face, adhering to Mistral’s licensing terms.

Pixtral 12B is built on Mistral’s text model, Nemo 12B. It can answer questions about numerous images of various sizes using either URLs or base64-encoded images. This capability places Pixtral in line with other advanced multimodal models.

Advanced Capabilities

Pixtral 12B stands out with its impressive ability to perform tasks such as captioning images and counting objects within photos. This opens up numerous applications for developers and researchers.

Currently, there are no working web demos of Pixtral 12B available, but Mistral has plans to offer testing on its chatbot and API platforms soon, according to Sophia Yang, head of developer relations at Mistral.

Data and Ethical Concerns

The specific image data used to develop Pixtral 12B remains unclear. Mistral, like many AI companies, often trains models on vast amounts of public data, which can include copyrighted content.

Disputes over ‘fair use’ rights are common, especially as copyright holders take legal action against companies that scrape public data without explicit permission. Mistral’s practices are part of a broader debate in the AI community.

A Significant Funding Boost

Pixtral 12B’s release follows Mistral raising $645 million in funding. This round was led by General Catalyst, and it valued the company at $6 billion. Such financial backing highlights the company’s strategic importance in the AI sector.

Mistral is seen as Europe’s answer to large American AI firms. Their strategy includes releasing free models while offering managed versions and consulting services to clients.

The Future of Multimodal Models

Multimodal models like Pixtral 12B could revolutionize various industries by integrating capabilities to process both text and images. This dual-functionality makes them highly versatile.

As more multimodal models are developed, the AI landscape will likely see significant shifts. Companies like Mistral are at the forefront of this evolution, pushing the boundaries of what AI can achieve.

Community and Developer Relations

Sophia Yang’s announcement on social media platforms like X highlights Mistral’s community engagement. The company actively updates its community about new releases and upcoming features.

Mistral’s approach to community engagement ensures that developers have the latest information and tools they need. This helps build a robust ecosystem around their models.

Licensing and Accessibility

Mistral’s standard license for Pixtral 12B allows free use for research and academic purposes. Commercial applications, however, require a paid license.

This licensing strategy ensures that the model remains accessible to a broad range of users while allowing the company to monetize its commercial applications.

Mistral’s Pixtral 12B marks a pivotal advancement in the field of AI, offering robust capabilities for both text and image processing.

As the company continues to innovate, Pixtral 12B sets a new standard for multimodal models.

About The Author

Emmanuel Kesse

See author's posts