MAZE
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
Maze
100Accuracy (Scale 8)
10
Maze
0.63Success Rate
10
maze_large (test)
70.5Success Rate
10
Maze MCR variant (100 held-out trajectories)
0.057D_policy
9
Maze hard (test)
85.3Accuracy
9
Maze (test)
99.9Accuracy
9
Maze 17^10 structured discrete
9.61Mean Score
9
Maze 100 held-out mazes
52.6Best Success Rate @ 3
8
Maze Hard
71.8RSR
8
Maze
1.03Mean Reward
8
Maze Line Seed
1.48Time (ms)
8
Maze Point Seed
1.32Time (ms)
8
MAZE
74.5EM
8
MAZE
84Pass@1 Accuracy
8
Maze OOD (test)
7.4L2 Error (%)
7
Maze ID (test)
3.7L2 Error
7
Maze OOD (test)
8.9Value Function L2 Error (%)
7
Maze ID (test)
2.7L2 Error (V)
7
Maze (OOD)
0.15Cosine Distance
7
Maze Structure Map
81.15Final Cumulative Win Rate (FW)
7
Maze 5^4 unstructured discrete
9.57Mean Performance
7
Maze 5^4 structured discrete
9.74Mean Score
7
Maze Hard
98Solvable Accuracy
7
Maze Easy
100Accuracy (Solvable)
7
MAZE
94.4Success Rate
6