Mathematical Reasoning on Hard problems
1,550Pass@1POPE (+ hard + guide)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| POPE (+ hard + guide)Training Mixture=Hard problems only2026.01 | 1,550 | 42.53 | |
| POPE (+ hard + guide + easy)Training Mixture=Hard (256) + Easy (256)2026.01 | 1,432 | 38.93 | |
| POPE (+ hard + guide + 1K easy problems)Training Mixture=Hard (256) + Easy (1K)2026.01 | 1,398 | 36.42 | |
| SFT (+ hard)Training Mixture=Hard problems only2026.01 | 1,355 | 32.89 | |
| SFT (+ hard + easy)Training Mixture=Hard (256) + Easy (256)2026.01 | 822 | 23.81 | |
| Prefix + rejection-sampled SFTTraining Mixture=Hard problems only2026.01 | 514 | 24.5 | |
| SFT (+ hard + 1K easy problems)Training Mixture=Hard (256) + Easy (1K)2026.01 | 224 | 25.12 | |
| Full-oracle SFTTraining Mixture=Hard problems only2026.01 | 200 | 12.37 | |
| Qwen3-4B-InstructType=Base model2026.01 | 57 | 7.42 |