Reward Modeling on RM-Bench Chat

78.5Accuracy

Gemini-2.5-Flash

Updated 5d ago

Evaluation Results

Method	Links
Gemini-2.5-Flash 2026.02		78.5
Gemini-2.5-Flash 2026.05		78.5
RM-R1-32B (Qwen-2.5-Inst) 2026.02		75.3
RM-R1-32B (Qwen-2.5-Inst) 2026.05		75.3
RM-R1-32B (DeepSeek-Dist) 2026.02		74.2
RM-R1-32B (DeepSeek-Dist) 2026.05		74.2
RRM-32B 2026.02		73.9
RRM-32B 2026.05		73.9
RM-R1-14B (Qwen-2.5-Inst) 2026.02		73.2
RM-R1-14B (Qwen-2.5-Inst) 2026.05		73.2
RM-R1-14B (DeepSeek-Dist) 2026.02		71.4
RM-R1-14B (DeepSeek-Dist) 2026.05		71.4
Rubric-ARM 2026.02		69.2
Rubric-ARM-voting@5 2026.02		69.1
Rubric-ARROW-voting@5 2026.05		68.6
API (Rubric+Judge) 2026.02		67.9
API Prompting (Rubric+Judge pairwise) 2026.05		67.9
Rubric-ARROW 2026.05		67.8
API (direct Judge) 2026.02		67.2
API Prompting (direct Judge) 2026.05		67.2
Rubric-RM-voting@5 2026.02		67
Rubric-RM-voting@5 2026.05		67
Rubric-ARROW w/o RL-voting@5 2026.05		66.9
Rubric-RM 2026.02		65.7
Rubric-RM 2026.05		65.7
Rubric-ARROW w/o RL 2026.05		64.8
RIFL 2026.05		64.8
RIFL-voting@5 2026.05		64.3
RM-R1-7B (Qwen-2.5-Inst) 2026.02		64.2
Qwen-3-8B (Rubric+Judge) 2026.02		64.2
RM-R1-7B (Qwen-2.5-Inst) 2026.05		64.2
Qwen-3-8B (Rubric+Judge pairwise) 2026.05		64.2
Claude-3.5-Sonnet 2026.02		62.5
Claude-3.5-Sonnet 2026.05		62.5
RM-R1-7B (DeepSeek-Dist) 2026.02		62.2
RM-R1-7B (DeepSeek-Dist) 2026.05		62.2
Qwen-3-8B (Rubric+Judge pointwise) 2026.05		61.2
RRM-7B 2026.02		59.9
RRM-7B 2026.05		59.9
JudgeLRM-7B 2026.02		55.4
JudgeLRM-7B 2026.05		55.4
API Prompting (Rubric+Judge pointwise) 2026.05		37.8