Mistral's first multimodal AI model takes on OpenAI's GPT-4o
What's the story
Mistral, a French artificial intelligence (AI) start-up, has launched its inaugural multimodal model named Pixtral 12B. It is capable of processing both images and text. With an impressive 12 billion parameters and a size of approximately 24GB, Pixtral 12B is designed to outperform models with fewer parameters in problem-solving tasks. Pixtral 12B is an extension of Mistral's text model, Nemo 12B and rivals OpenAI's GPT-4o.
Model features
Pixtral 12B's capabilities
Pixtral 12B can respond to queries about an unlimited number of images of any size, provided either through picture URLs or images encoded using base64. The model is expected to perform tasks such as captioning images and counting objects in a photo, similar to other multimodal models like Anthropic's Claude family and OpenAI's GPT-4o.
Distribution
Availability and licensing details
Pixtral 12B is accessible via a torrent link on GitHub and Hugging Face, platforms dedicated to AI and machine learning development. Users can download, fine-tune, and utilize the model under Mistral's standard license. This license mandates a paid license for commercial applications, but does not require one for research or academic purposes.
Future testing
Pixtral 12B to be tested on Mistral's platforms
Sophia Yang, head of Mistral's developer relations, announced that Pixtral 12B will soon be offered for testing on Mistral's chatbot and API-serving platforms, Le Chat and Le Platforme. However, at the time of publication, no working web demos were available for Pixtral 12B. The specific image data used by Mistral to develop this model remains undisclosed.
Company growth
Mistral's funding and market strategy
The launch of Pixtral 12B follows Mistral's successful $645 million funding round led by General Catalyst, which valued the firm at $6 billion. Though its just over a year old, Mistral is already being hailed as Europe's answer to OpenAI. The company's strategy includes the release of free "open" models, charging for managed versions of these models, and offering consulting services to corporate clients.