World Embedding Benchmark

Researchers introduced the World Embedding Benchmark, a dataset of 8,000 controlled simulation cases drawn from 80 families spanning fluid mechanics, to measure how video representations encode physical information. The benchmark targets a gap in understanding of physical fidelity in world models and video generation, where how video representations encode physical information remains less understood.

Physical fidelity has received increasing attention in world models and video generation, yet how video representations encode physical information remains less understood. We introduce the World Embedding Benchmark, comprising 8,000 controlled simulation cases from 80 families spanning fluid mechan