Mathematical Reasoning on AIME 24 (Accuracy Avg@8)
82.1Accuracy (Avg@8)CERL full
Evaluation Results
| Method | Links | |
|---|---|---|
| CERL fullTraining=SFT+RL2026.05 | 82.1 | |
| Correctness-only RLTraining=RL2026.05 | 79.6 | |
| HaloTraining=SFT+RL2026.05 | 79.5 | |
| Long-answer SFTTraining=SFT2026.05 | 78.4 | |
| Fixed-ratio erasure + free answerTraining=None2026.05 | 76.4 | |
| Base Qwen3-8BTraining=N/A2026.05 | 75.9 | |
| Length-penalty RLTraining=RL2026.05 | 75.1 | |
| TokenSkipTraining=SFT2026.05 | 72 |