Mistral, a French AI startup, has released its first multimodal AI model called Pixtral 12B. The model can process both images and text, allowing users to ask questions about the subjects in images. Pixtral 12B is built on Mistral’s existing text-based model, Nemo 12B.
It has 12 billion parameters and is approximately 24GB in size. The model is designed to perform tasks such as captioning images, identifying objects, and answering image-related queries. The model is available for free under the Apache 2.0 license, which allows users to use, modify, or commercialize it without restrictions.
Developers can download the model from GitHub and Hugging Face, although functional web demos are not yet live.
Mistral integrates Pixtral into services
According to Sophia Yang, head of developer relations at Mistral, Pixtral 12B will soon be integrated into the company’s chatbot, Le Chat, and API platform, La Platforme.
The release of Pixtral 12B comes after a significant funding round for Mistral, which pushed its valuation to $6 billion. Microsoft is among the company’s backers, positioning Mistral as Europe’s response to OpenAI. However, questions remain over the data sources used to train these models.
Like many AI firms, Mistral likely utilized vast amounts of publicly available web data, a practice that has sparked lawsuits in the tech industry. Mistral continues to make strides in the AI sector, aiming to further integrate Pixtral 12B into various applications and services. The company’s approach has been to offer free, open models while charging for managed versions and consulting services for corporate clients.
