Categories Companies Research

Loading the SOTA2 catalog…

Explore

Categories Companies Research

Contribute

Join SOTA2 Contact

SOTA2

© 2026 SOTA2

Build in Bellevue, USA

WizWand is now SOTA2 Research

Papers Benchmarks Datasets Tasks

Reward Modeling research benchmarks – Page 6 · SOTA2 Research

Reward Modeling

Benchmarks

Dataset NameSOTA MethodSort

Dataset NameSOTA methodMetricTrendResultsLast Updated

Llama-3.1-Nemotron-70B-Reward

93.8Clarity

9

May 19, 2026

PsyCoPref v1.0 (test)

PsyCo-Llama3-3B-Reward

98.1Accuracy

9

Apr 14, 2026

RewardBench mathprm

gpt-4o-2024-05-13

72.5Score

9

Mar 19, 2026

HHH-Alignment Reversed

86.2Accuracy

9

Feb 26, 2026

HHH-Alignment Standard

91.8Accuracy

9

Feb 26, 2026

SafeRLHF Reversed

88.4Accuracy

9

Feb 26, 2026

SafeRLHF Standard

89.8Accuracy

9

Feb 26, 2026

Overall Performance (test)

80.2Overall

9

Feb 26, 2026

RewardBench 2 (test)

76.3RWBench2 Score

9

Feb 26, 2026

59.7Instruction Following (IF) Score

8

Jun 23, 2026

48.9SRCC (Alignment)

8

Jun 11, 2026

VideoScore-Bench v2

46Accuracy (Alignment)

8

Jun 11, 2026

VideoFeedback (test)

82.62Visual Score

8

May 28, 2026

PPE-Preference 1k

51.7Positional Consistency

8

Apr 10, 2026

73.5Positional Consistency

8

Apr 10, 2026

56.3Positional Consistency Score

8

Apr 10, 2026

UltraFeedback Cleaned

92.36Total Score

8

Mar 6, 2026

Aggregate RewardBench PPE RMB RM-Bench JudgeBench

Skywork-Reward-V2

82.8Average Score

8

Mar 4, 2026

Skywork-Reward-V2

76.9PPE Human Preference

8

Mar 4, 2026

58.4Accuracy

8

Feb 26, 2026

HHH-Alignment (OOD)

79.8Accuracy

8

Feb 26, 2026

Unified-Feedback (ID)

GRM w/ dpo-noref

73.9Accuracy

8

Feb 26, 2026

MT-Bench OOD (test)

73Score

8

Feb 26, 2026

HHH-Alignment OOD (test)

78.7Score

8

Feb 26, 2026

Unified-Feedback ID (test)

71.5Reward Score

8

Feb 26, 2026

Page 6 of 10

Popular datasets

RewardBench RM-Bench UltraFeedback PersonalLLM PPE

Follow for updates

Join SOTA2 Contact