VSI-Bench (full name Visual-Spatial Intelligence Benchmark) is a visual spatial intelligence benchmark test set launched by Fei-Fei Li, Sai-Ning Xie and their research team in 2024. It aims to evaluate the ability of multimodal large language models (MLLMs) in spatial cognition and understanding. The related paper results are "Thinking in Space: How Multimodal Large Language Models See, Remember, and Recall Spaces". The dataset contains more than 5k question-answer pairs, covering nearly 290 real indoor scene videos, involving a variety of environments such as homes, offices, and factories, and covering multiple aspects of problems such as object recognition, positional relationships, and action prediction. This diverse data structure not only helps train more robust models, but also provides developers with rich resources for algorithm verification and optimization.