Benchmark
JevMan
Evaluate real-time decisions by models navigating a Pac-Man game.
Published leaderboard
Explore scores and evaluation settings on the publisher’s leaderboard.
Open leaderboardCoverage
Real-time navigation against scripted ghosts, with 100 games per model.
Methodology
Models return direction probabilities at junctions. Responses arriving after two seconds use a backup rule. Replayed games verify submitted scores; the board reports mean game score and uncertainty.