Pairwise Reference Alignment on RewardBench benchmark leaderboard · SOTA2 Research

Pairwise Reference Alignment on RewardBench

0.7705A^MQwen2.5-7B-Instruct

Evaluation Results

Method	Links
Qwen2.5-7B-Instruct2026.05		0.7705	0.759	0.35	0.3228
Qwen2.5-7B2026.05		0.7262	0.7137	0.2005	0.1817
Qwen2.5-3B-Instruct2026.05		0.7232	0.7113	0.2665	0.2362
Qwen2.5-3B2026.05		0.6928	0.6799	0.1648	0.1447
Qwen2.5-1.5B-Instruct2026.05		0.6715	0.6582	0.1585	0.1349
Qwen2.5-1.5B2026.05		0.6652	0.6527	0.1388	0.1168
Qwen2.5-0.5B-Instruct2026.05		0.625	0.6119	0.1177	0.0926
Qwen2.5-0.5B2026.05		0.6148	0.6018	0.0944	0.0707