Reward Model Evaluation on Reward Bench 2 (test) benchmark leaderboard · SOTA2 Research

Reward Model Evaluation on Reward Bench 2 (test)

0.451RB2 Factuality MAEDistribution-Calibrated Aggregation

Evaluation Results

Method
Distribution-Calibrated Aggregation2025.12	0.451	0.287	0.285	0.414	0.285	0.081
Distribution-Calibrated Aggregation2025.12	0.487	0.332	0.306	0.451	0.319	0.094
USC2025.12	0.573	0.423	0.415	0.53	0.405	0.221
USC2025.12	0.575	0.424	0.41	0.574	0.407	0.226
GSC2025.12	0.591	0.441	0.45	0.524	0.414	0.197
GSC2025.12	0.599	0.439	0.427	0.597	0.406	0.208
SC2025.12	0.615	0.394	0.36	0.498	0.373	0.155
SC2025.12	0.647	0.403	0.384	0.552	0.402	0.158
CI-SC2025.12	0.67	0.415	0.385	0.57	0.405	0.165
CI-SC2025.12	0.675	0.415	0.391	0.551	0.412	0.178
Soft-SC2025.12	0.681	0.397	0.4	0.581	0.406	0.177
Soft-SC2025.12	0.711	0.37	0.372	0.603	0.409	0.177