Mathematical Reasoning on Macro-average
36.6Avg@8DAPO w/ PR-FKL
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DAPO w/ PR-FKLModel Backbone=Qwen3-8B-Base2026.05 | 36.6 | 54.74 | |
| DAPO w/ PR-RKLModel Backbone=Qwen3-8B-Base2026.05 | 36.02 | 50.12 | |
| DAPO w/ RRModel Backbone=Qwen3-8B-Base2026.05 | 35.93 | 52.66 | |
| DAPOModel Backbone=Qwen3-8B-Base2026.05 | 34.27 | 51.6 | |
| DAPO w/ PR-FKLModel Backbone=Qwen3-4B-Base2026.05 | 31.46 | 46.05 | |
| DAPO w/ PR-RKLModel Backbone=Qwen3-4B-Base2026.05 | 30.98 | 45.41 | |
| DAPO w/ RRModel Backbone=Qwen3-4B-Base2026.05 | 30.86 | 44.7 | |
| DAPOModel Backbone=Qwen3-4B-Base2026.05 | 29.4 | 43.27 | |
| BaseModel Backbone=Qwen3-8B-Base2026.05 | 17.86 | 35.03 | |
| BaseModel Backbone=Qwen3-4B-Base2026.05 | 17.65 | 35.93 |