MAZE
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
Maze Hard
81Exact Match Rate
6
Maze Med
99.5Exact Match Rate
6
Maze Easy
100Exact Match Accuracy
6
Maze Overall
90.3Exact Match Percentage
6
Maze
50Total Sampled Points
6
maze-32-32-4 (# agents: 30)
4.75UA Conflicts
6
maze 32-32-2 (# agents: 30)
7.59UA Conflicts
6
Maze
3,518Token Cost
6
Maze 4× OOD
4.5Exact Solve Rate
5
Maze (2× OOD)
98.1Exact Solve Rate
5
Maze T ∈ [21 : 60]
0.054VRMSE
4
Maze
0.035VRMSE
4
Maze 2
100Success Rate (SR)
4
Maze 1
100Success Rate (SR)
4
Maze (Hard)
83.1Success Rate
4
Maze |A| = 400 (test)
2,417Throughput
4
Maze 17^10 unstructured discrete
9.25Mean Score
4
maze-10 POMDP PRISM format (original enlarged)
8.86Value (IQM)
4
MAZE
85.5Pass@1
4
Maze (24, 124) (test)
100Accuracy
4
Medium Maze
-0.88NLL
4
Maze free-response (OOD)
85.71Visual Reward (rv)
3
Maze
0.032Mean Path Length
3
Maze unseen (test)
77Success Rate
3
Maze
21.5Accuracy
3