A cutting-edge benchmark to evaluate the predictive reasoning capabilities of Multimodal Large Language Models.