General Reasoning on MMLU (Pass@1, #Tok)
49.39Pass@1 AccuracySAS
Evaluation results
| Method | Links | ||
|---|---|---|---|
| SASmax context length=4K, ratio r=0.32026.04 | 49.39 | 876 | |
| L1-Maxmax context length=4K2026.04 | 48.94 | 900 | |
| ThinkPrune-4kmax context length=4K2026.04 | 48.91 | 973 | |
| GRPO-4Kmax context length=4K2026.04 | 48.73 | 950 | |
| LAPO-Imax context length=4K2026.04 | 48.71 | 1,207 | |
| DeepScaleRmax context length=4K2026.04 | 47.99 | 1,454 |