​Mistral launches the most powerful open source multi-modal model Pixtral Large. Upgrading Le Chat can directly call Flux Pro

French artificial intelligence startup Mistral AI has announced a series of new features for its Le Chat AI assistant, including integrated web search, image generation, and the newly launched Pixtral Large model.

Le Chat function upgrade

Users can now directly access real-time web content through Le Chat and easily obtain the information they need. At the same time, with the help of Black Forest Labs' Flux Pro model, users can also generate high-quality images to meet a variety of creative needs.

In addition to web search and image generation, Le Chat also introduces a canvas interface that allows users to edit generated content directly within the chat window. This feature enables users to write documents, create presentations, and edit code without having to regenerate responses, greatly improving work efficiency.

Introducing Pixtral Large model

The Pixtral Large model launched by Mistral AI performs very well in visual tasks. This model is built on Mistral Large2 and has achieved excellent results in multiple industry benchmarks.

For example, in the MathVista mathematical reasoning test, Pixtral Large scored 69.4%, surpassing other competitors such as GPT-4o and Gemini1.5Pro.

At the same time, the model has also been recognized for its ability to analyze charts and complex documents, capable of processing a variety of information including graphs, tables and formulas.

The Pixtral Large model combines a 123 billion parameter multi-modal decoder with a 1 billion parameter visual encoder, and can process up to 128 high-resolution images simultaneously, with a maximum context window of 30K.

This makes it excellent at document analysis and complex image processing. Mistral AI stated that Pixtral Large will also provide both academic and commercial licenses on the Hugging Face platform to facilitate research and application by different users.

In addition, Mistral AI has updated its Mistral Large language model to improve the accuracy of long context understanding and function calls.

The updated model will be available through Mistral's API and will soon be available on Google Cloud and Microsoft Azure.

Pixtral Large paper entrance: https://arxiv.org/abs/2410.07073

Model page: https://huggingface.co/mistralai/Pixtral-Large-Instruct-2411

Le Chat entrance: https://auth.mistral.ai/ui/login?flow=b3e9d399-afc8-497b-8f8d-99900b447c08

API entrance: https://docs.mistral.ai/api/