RewardBench

Benchmarks

Task Name	Dataset Name	SOTA Result
Reward Modeling	RewardBench	Accuracy97.8	166
Reward Modeling	RewardBench	Chat Score99.4	146
Reward Modeling	RewardBench v1.0 (test)	Average Score0.978	89
Reward Modeling	RewardBench Focus 2	Accuracy90.3	82
Reward Modeling	RewardBench v2	Accuracy92.1	72
Reward Modeling	RewardBench Precise IF 2	Accuracy57.5	70
Reward Modeling	RewardBench Average 2	Accuracy39.7	52
Reward Modeling	RewardBench Math 2	Accuracy35.7	52
Reward Modeling	RewardBench v2 (test)	Average Score86.5	42
LLM-as-a-Judge	RewardBench 1.0 (test)	Rstd0.54	36
LLM-as-a-Judge Evaluation Consistency	RewardBench	Kappa83.25	36
Multimodal Reward Modeling	RewardBench Multimodal	Safety Score99.6	31
Reward Modeling	RewardBench 2	Accuracy89.5	30
Multimodal Reward Modeling	Multimodal RewardBench	Accuracy88.79	30
Pair-wise comparison	RewardBench	Accuracy93.7	29
Reward Modeling	RewardBench v1	Accuracy95.5	28
Reward Modeling	RewardBench (test)	RWBench0.933	25
Uncertainty Calibration	RewardBench	Kuiper0.009	24
Reward Modeling Evaluation	RewardBench2 (test)	Accuracy82.9	20
Reward Modeling	RewardBench 2	L-Acc93.4	20
Reward Modeling	RewardBench unified-feedback (test)	Average Score84	20
Multi-modal Preference Evaluation	MM-RewardBench	Accuracy72.9	19
Reward Modeling	RewardBench Chat	Accuracy96.4	18
Multimodal Reward Modeling	MM-RLHF-RewardBench	Pairwise Accuracy92.4	18
Reward Modeling	RewardBench 1k	Positional Consistency84.9	16

Showing 25 of 47 rows