AI startup Mistral from France announced the launch of a new open multimodal model Pixtral Large, which, according to the announcement on the company’s official website, has 124 billion parameters and is capable of processing images, diagrams and various types of documents.
AI startup Mistral from France announced the launch of a new open multimodal model Pixtral Large, which, according to the announcement on the company’s official website, has 124 billion parameters and is capable of processing images, diagrams and various types of documents.
«Pixtral Large is suitable for workflows related to finding and retrieving information, automating tasks and working with complex documents, making it useful for enterprise purposes», — Mistral’s offsite said in a post dated November 18.
Pixtral Large is based on the company’s flagship Mistral Large 2 development.
For scientific purposes, Pixtral Large is available under the Mistral Research License, and for commercial use under the Mistral Commercial License.
The Mistral company assures that the MathVista test showed that the Pixtral Large model outperformed the competition and showed a result of 69,4%.
Mistral also announced that they have improved the flagship Mistral Large 2 itself: it now processes large text arrays better, received a hint system and clearer functionality.
At the end of May, we reported that French AI startup Mistral, which was backed by Microsoft and was valued at $6 billion at the time, released its first generative AI model for coding called Codestral.
