Tag
Introduces WorldBench, a visually diverse multimodal reasoning benchmark that reveals significant limitations in current multimodal large language models' visual understanding.