Reward Modeling Assessment on IF-RewardBench benchmark leaderboard · SOTA2 Research

Reward Modeling Assessment on IF-RewardBench

0.619Single-Turn ScoreSkill-RM

Evaluation Results

Method	Links
Skill-RM2026.06		0.619	0.54	0.413	0.524
Gemini-3-Flash2026.06		0.589	0.46	0.489	0.513
GPT-5-mini2026.06		0.521	0.438	0.41	0.456
Qwen3.5-27B2026.06		0.507	0.44	0.287	0.411
DeepSeek-V3.22026.06		0.397	0.257	0.208	0.288
GLM-4.62026.06		0.359	0.263	0.189	0.27
Skywork-V2-Llama3.1-8B2026.06		0.153	0.205	0.039	0.133
RRM-32B2026.06		0.144	0.069	0.004	0.072
LMUnit-Qwen2.5-72B2026.06		0.126	0.108	0.009	0.081
RM-R1-Qwen32B2026.06		0.114	0.075	-0.032	0.052
Llama3.1-70B-RM2026.06		0.109	0.208	0.054	0.124
Qwen2.5-Math-RM2026.06		0.056	0.092	0.002	0.05
Qwen2.5-72B-Instruct2026.06		0.052	0.097	-0.003	0.048