Tag
MuseBench is a comprehensive benchmark introduced to evaluate multimodal large language models on nuanced, intent-level understanding of audiovisual arts, revealing that even the best model achieves only 48.29% accuracy compared to 87.18% for human experts.