Mathematical Reasoning on AIME 24 (Accuracy, Length, #Relf./Ans.)
61.67AccuracyHybrid
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| HybridModel=Qwen3-4B, Mode=Think2026.04 | 61.67 | 11,595.51 | 45.85 | |
| PLEModel=Qwen3-4B, Mode=Think2026.04 | 60 | 31,733.18 | 7.02 | |
| PLEModel=Qwen2.5-7B, Mode=Think2026.04 | 40 | 11,794.06 | 62.76 | |
| PLEModel=Qwen3-4B, Mode=No-think2026.04 | 40 | 5,597 | 0.39 | |
| SFT-onlyModel=Qwen2.5-7B, Mode=Think2026.04 | 30 | 12,517.06 | 76.44 | |
| PLEModel=Qwen2.5-7B, Mode=No-think2026.04 | 26 | 1,328.6 | 0.02 | |
| HybridModel=Qwen3-4B, Mode=No-think2026.04 | 20.67 | 4,636.37 | 2.54 | |
| InstructModel=Qwen2.5-7B, Mode=No-think2026.04 | 6.67 | 1,729.22 | 0 | |
| SFT-onlyModel=Qwen2.5-7B, Mode=No-think2026.04 | 3.33 | 1,037.2 | 0 |