ResearchBenchmarksInference Speed Evaluation on long-term video generation evaluation 561-frameFollow1.71Gen. Time (s/frame)VWM1.6462.0782.512.942Feb 26, 2026Evaluation ResultsMethodMethodLinksGen. Time (s/frame)Avg. 3D Recon. Time (s/frame)Total Time (s/frame)VWM2026.021.710.42.11UCM2026.022.40.362.76C-a-M2026.022.86—2.86VMem2026.023.310.43.71