List your company

ResearchFormerly WizWand

Reward Modeling

Benchmarks

DatasetSOTA methodMetricTrendResultsLast updated

ARMBench-VL ours (test)

67.6FG Score

7

Feb 26, 2026

57.9Accuracy

7

Feb 26, 2026

Arabic preference (test)

RM-Distiller-Qwen2.5-3B-Instruct

85.4Accuracy

7

Feb 26, 2026

0.75Accuracy

7

Feb 26, 2026

RewardBench External Evaluation

93.3Chat Score

6

Jul 3, 2026

71.6Accuracy

6

Jun 11, 2026

RewardBench and RM-Bench Aggregate 2

90.3Average Score

6

Jun 9, 2026

SummEval (test)

0.044MSE (Table)

6

Jun 2, 2026

RewardBench RLHFlow source Chat Chat-Hard Safety Reasoning Total 1.0 (train)

Difficulty-Based Preference Data Selection

80.62Chat Score

6

May 19, 2026

RewardBench UltraFeedback training source Chat Chat-Hard Safety Reasoning Total 1.0

Difficulty-Based Preference Data Selection

80.98Chat Score

6

May 19, 2026

Anthropic/hh-rlhf HH-helpful core250

0.292Delta RM

6

May 12, 2026

D_dish real policy rollouts (test)

0.67Rollout ρ

6

Apr 28, 2026

0.013Demo Loss

6

Apr 28, 2026

ManiRewardBench Single-arm YAM

0.945Mean VOC

6

Feb 26, 2026

ManiRewardBench Bimanual YAM

94.7Mean VOC

6

Feb 26, 2026

ManiRewardBench Franka

94.2Mean VOC

6

Feb 26, 2026

ManiRewardBench Lerobot

0.954Mean VOC

6

Feb 26, 2026

PersonalRewardBench (test)

65.21Macro Accuracy

6

Feb 26, 2026

79.42Accuracy

5

Jul 2, 2026

0.204Test MSE (Table)

5

Jun 2, 2026

Internal preference benchmark (hold-out)

Skywork-Reward-V2-Llama-3.1-8B

81.2Human Agreement

5

Mar 4, 2026

UltraMedical (test)

PairRM-LLaMA3-8B

95.8Easy Score

5

Feb 26, 2026

Anthropic HH (test)

Dahoas/gptj-rm-static

68.49Accuracy

5

Feb 26, 2026

Meta Safety (test)

64.5Accuracy

5

Feb 26, 2026

Meta Helpfulness (test)

63.2Accuracy

5

Feb 26, 2026