Mathematical Reasoning on MATH-500 (Acc, Tok, ReL Metrics)
90AccuracyOurs
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| OursBackbone=Qwen2.5-7B2026.02 | 90 | 4,196 | 140.6 | |
| Ours (Base)Backbone=Qwen2.5-7B2026.02 | 87.4 | 4,253 | 161.6 | |
| INFTYTHINKBackbone=Qwen2.5-7B2026.02 | 86.2 | 5,050 | 184.8 | |
| LightThinkerBackbone=Qwen2.5-7B2026.02 | 84.8 | 4,321 | 145.2 | |
| SepLLMBackbone=Qwen2.5-7B2026.02 | 84.2 | 4,244 | 144.3 | |
| H2OBackbone=Qwen2.5-7B2026.02 | 82.6 | 4,389 | 147.1 | |
| OursBackbone=Qwen2.5-1.5B2026.02 | 81.2 | 3,812 | 125.8 | |
| Ours (Base)Backbone=Qwen2.5-1.5B2026.02 | 78.8 | 3,958 | 139.4 | |
| INFTYTHINKBackbone=Qwen2.5-1.5B2026.02 | 78 | 4,750 | 170.2 | |
| LightThinkerBackbone=Qwen2.5-1.5B2026.02 | 77.6 | 4,060 | 134.8 | |
| SepLLMBackbone=Qwen2.5-1.5B2026.02 | 76 | 3,890 | 130.3 | |
| H2OBackbone=Qwen2.5-1.5B2026.02 | 75.4 | 3,835 | 128.5 |