Reward Modeling on RewardBench Chat

96.4Accuracy

Claude-3.5-Sonnet

Updated 5d ago

Evaluation Results

Method	Links
Claude-3.5-Sonnet 2026.02		96.4
Claude-3.5-Sonnet 2026.05		96.4
RM-R1-32B (Qwen-2.5-Inst) 2026.02		95.3
RM-R1-32B (DeepSeek-Dist) 2026.02		95.3
RM-R1-32B (Qwen-2.5-Inst) 2026.05		95.3
RM-R1-32B (DeepSeek-Dist) 2026.05		95.3
Gemini-2.5-Flash 2026.02		95
Gemini-2.5-Flash 2026.05		95
RRM-32B 2026.02		94.7
RRM-32B 2026.05		94.7
JudgeLRM-7B 2026.02		92.1
JudgeLRM-7B 2026.05		92.1
Rubric-ARROW-voting@5 2026.05		90.8
RM-R1-14B (DeepSeek-Dist) 2026.02		90.3
Rubric-ARM-voting@5 2026.02		90.3
RM-R1-14B (DeepSeek-Dist) 2026.05		90.3
Rubric-RM-voting@5 2026.02		89.9
Rubric-RM-voting@5 2026.05		89.9
RIFL-voting@5 2026.05		89.7
API (direct Judge) 2026.02		89.6
API Prompting (direct Judge) 2026.05		89.6
RIFL 2026.05		89.6
Rubric-ARM 2026.02		89.4
Rubric-ARROW 2026.05		89.1
Rubric-RM 2026.02		88.2
Rubric-RM 2026.05		88.2
Rubric-ARROW w/o RL-voting@5 2026.05		87.7
Rubric-ARROW w/o RL 2026.05		87.4
RM-R1-7B (DeepSeek-Dist) 2026.02		85.3
RM-R1-7B (DeepSeek-Dist) 2026.05		85.3
RM-R1-7B (Qwen-2.5-Inst) 2026.02		83
RM-R1-7B (Qwen-2.5-Inst) 2026.05		83
API (Rubric+Judge) 2026.02		79.6
API Prompting (Rubric+Judge pairwise) 2026.05		79.6
RRM-7B 2026.02		77.7
RRM-7B 2026.05		77.7
Qwen-3-8B (Rubric+Judge) 2026.02		73.9
Qwen-3-8B (Rubric+Judge pairwise) 2026.05		73.9
RM-R1-14B (Qwen-2.5-Inst) 2026.02		73.5
RM-R1-14B (Qwen-2.5-Inst) 2026.05		73.5
Qwen-3-8B (Rubric+Judge pointwise) 2026.05		69.3
API Prompting (Rubric+Judge pointwise) 2026.05		51.7