Reward Bench
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
Reward Bench Factuality 2
56.6Pairwise Accuracy
64
Reward Bench Math
0.305EF
52
Reward Bench safety subset response perturbations 2
-0.629LE Score
26
Reward Bench safety subset prompt perturbations 2
-0.18EF
26
Reward Bench V2
83.44Accuracy
22
Reward Bench Prior Sets
78.2Prior Sets Score
17
Reward Bench 2 (test)
0.451RB2 Factuality MAE
12
Reward Bench Ties 2
91.8Pairwise Accuracy
12
Reward Bench Safety 2
72.3Pairwise Accuracy
12
Reward Bench Math 2
72.3Pairwise Accuracy
12
Reward-Bench
84.79Agreement
12