Reward Model Evaluation on internally annotated (test)
0.762PLCCGDSO
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| GDSOModel Size=Qwen3.5-27B, Scoring Based on Reasoning=✓2026.06 | 0.762 | 0.7132 | 89.56 | 98.85 | |
| RISDModel Size=Qwen3.5-9B, Scoring Based on Reasoning=Internalized2026.06 | 0.7391 | 0.6882 | 88.64 | 98.01 | |
| GRPOModel Size=Qwen3.5-27B, Scoring Based on Reasoning=✓2026.06 | 0.72 | 0.6832 | 86.04 | 98.27 | |
| RewardDanceModel Size=Qwen3.5-27B, Scoring Based on Reasoning=×2026.06 | 0.6667 | 0.6207 | 84.25 | 97.06 | |
| SFTModel Size=Qwen3.5-27B, Scoring Based on Reasoning=×2026.06 | 0.6458 | 0.5914 | 81.35 | 96.44 | |
| GDSOModel Size=Qwen3.5-9B, Scoring Based on Reasoning=✓2026.06 | 0.6341 | 0.5665 | 83.95 | 95.99 | |
| Zero-shotModel Size=Qwen3.5-27B, Scoring Based on Reasoning=×2026.06 | 0.6301 | 0.5816 | 74.38 | 95.38 | |
| GRPOModel Size=Qwen3.5-9B, Scoring Based on Reasoning=✓2026.06 | 0.534 | 0.5072 | 77.03 | 90.76 | |
| SFTModel Size=Qwen3.5-9B, Scoring Based on Reasoning=×2026.06 | 0.5296 | 0.4942 | 74.59 | 84.01 | |
| RewardDanceModel Size=Qwen3.5-9B, Scoring Based on Reasoning=×2026.06 | 0.5182 | 0.4338 | 78.17 | 89.72 | |
| Zero-shotModel Size=Qwen3.5-9B, Scoring Based on Reasoning=×2026.06 | 0.3411 | 0.3167 | 65.63 | 75.01 |