ResearchBenchmarksHuman Preference Evaluation on Minecraft Human Preference Study (model-vs-model trials)Follow290WinsMD4209.92230.71251.5272.29Apr 22, 2026Evaluation ResultsMethodMethodLinksWinsAppearancesPreference RatePreference Rate (95% CI Lower Bound)MD4version=p2version=p22026.0429050557.453.1DDPMversion=p2version=p22026.0428253053.249MD4version=p4version=p42026.0425551749.345Real2026.0421352840.336.2