WRBench Leaderboard

Model-level profile for Current World Models Lack a Persistent State Core. WRBench separates camera execution, visible event consistency, re-observation support, and returned-state closure instead of collapsing evidence into one opaque score.

26models
6D1-D6 axes
4condition groups
2input modes
Scores shown Avg D1 camera D2-D4 visible Support D5-D6 returned
Loading data...
Compare