Pixtral Large

Mistral AI announced Pixtral Large, a 124B parameter open-weights multimodal model built upon Mistral Large 2. It achieved state-of-the-art performance on multimodal benchmarks like MathVista and DocVQA, surpassing competitors such as GPT-4o and Gemini-1.5 Pro. The model was designed for complex reasoning on documents, charts, and natural images, featuring a 128K context window. This specific model is now deprecated and has been superseded by the company's more recent and powerful vision models.