Current World Models Lack a Persistent State Core
Paper • 2606.20545 • Published • 16
WRBench: Current World Models Lack a Persistent State Core
Official WRBench paper, public datasets, benchmark videos, human annotations, and interactive leaderboard.
Note WRBench paper page
WRBench/wrbench-natural25
Viewer • Updated • 475 • 112Note Natural-25 prompts, variants, and first frames
WRBench/wrbench-results
Viewer • Updated • 26 • 66Note 26-model aggregate diagnostic scores
WRBench/wrbench-human-annotations
Viewer • Updated • 3.78k • 56Note Human pairwise annotation verdicts
WRBench/wrbench-videos
Viewer • Updated • 11.1k • 12.3kNote 26-model / 11,100 benchmark videos with per-video scores
WRBench Leaderboard
🎬2Diagnostic D1-D6 leaderboard for world models
Note Interactive D1-D6 leaderboard