November 19th.Mistral AI Corp. made an announcement yesterday (Nov. 18) about a newMultimodality AI Models Pixtral Large.The model has 124 billion parameters, is based on Mistral Large 2, and is mainly used for processing text and images.
Pixtral Large is now available under the Mistral Research License and Commercial License for research, education, and commercial use.
Pixtral Large is the second model in the Mistral AI multimodal family.IT House cites an official press release that the model performed well in standard multimodal benchmarks such as MathVista, DocVQA and VQAv2, especially in MathVista where it achieved an accuracy of 69.4%, outperforming all competitors.
Pixtral Large also outperformed GPT-4o and Gemini-1.5 Pro in both the ChartQA and DocVQA tests.
Equipped with a 123B multimodal decoder and a 1B visual encoder, the model supports a 128K context window and is capable of processing at least 30 high-resolution images.Pixtral Large excels not only with visual data, but also with complex reasoning and graphical understanding.
