MMWorld is a new benchmark designed to evaluate the capabilities of Multimodal Large Language Models (MLLMs) in understanding and reasoning about complex real-world dynamics through video content. The benchmark covers seven broad disciplines and 69 subdisciplines, focusing on multi-faceted reasoning beyond perception, such as explanation, counterfactual thinking, future prediction, and domain expertise. MMWorld consists of a human-annotated dataset and a synthetic dataset. The human-annotated dataset includes 1,559 question-answer pairs and video captions, while the synthetic dataset is used to analyze MLLMs' perception within single modalities. The benchmark evaluates 12 MLLMs, including both open-source and proprietary models, and reveals significant challenges for existing models, with even the best performer achieving only 52.3% accuracy. The study also highlights differences in skill sets between MLLMs and humans, providing insights into their capabilities and limitations. MMWorld aims to serve as a crucial step towards comprehensive world model evaluation in videos.MMWorld is a new benchmark designed to evaluate the capabilities of Multimodal Large Language Models (MLLMs) in understanding and reasoning about complex real-world dynamics through video content. The benchmark covers seven broad disciplines and 69 subdisciplines, focusing on multi-faceted reasoning beyond perception, such as explanation, counterfactual thinking, future prediction, and domain expertise. MMWorld consists of a human-annotated dataset and a synthetic dataset. The human-annotated dataset includes 1,559 question-answer pairs and video captions, while the synthetic dataset is used to analyze MLLMs' perception within single modalities. The benchmark evaluates 12 MLLMs, including both open-source and proprietary models, and reveals significant challenges for existing models, with even the best performer achieving only 52.3% accuracy. The study also highlights differences in skill sets between MLLMs and humans, providing insights into their capabilities and limitations. MMWorld aims to serve as a crucial step towards comprehensive world model evaluation in videos.