RewardBench
Benchmarks
TaskDatasetSOTA resultTrendResults
RewardBench
0.009Kuiper
24
RewardBench 2
88.2Factuality
21
RewardBench 2
74.5Accuracy
20
RewardBench
0.542Spearman ρ
20
RewardBench MM-RLHF
95.74MCQ Score
20
RewardBench2 (test)
82.9Accuracy
20
RewardBench 2
93.4L-Acc
20
RewardBench unified-feedback (test)
84Average Score
20
MM-RewardBench
72.9Accuracy
19
MM-RLHF-RewardBench
92.4Pairwise Accuracy
18
RewardBench 1k
84.9Positional Consistency
16
RewardBench
100Coverage
16
RewardBench2
85Score
15
RewardBench Chat-Hard
89.4Chat-Hard Score
15
RewardBench English
91.01Accuracy
14
m-RewardBench Avg. 23 langs
91.37Accuracy
14
IF-RewardBench
0.619Single-Turn Score
13
VL-RewardBench, Multimodal RewardBench, and MM-RLHF-RewardBench Aggregate
84.1Accuracy
13
RewardBench
93.9Accuracy
12
RewardBench (CMI-Pref)
78.2Pairwise Accuracy
11
CMI-RewardBench (PAM)
0.698SRCC
11
RewardBench latest (full)
93.6Average Score
11
RewardBench 2
68.42Factuality
10
RewardBench 2
73.9Accuracy
10
RewardBench Multimodal
80.1Accuracy
10