Pixtral Large

Mistral AI has released Pixtral Large, a 124B open-weights multimodal model built on top of Mistral Large 2. The model demonstrates frontier-level image understanding for documents, charts, and natural images while maintaining the text performance of its predecessor. According to benchmarks, Pixtral Large achieves state-of-the-art results on MathVista and DocVQA, outperforming models like GPT-4o and Gemini-1.5 Pro. The model is available via API, on le Chat, and for download under research and commercial licenses.