Mathematical Reasoning on MATH 500 (Pass@1, Token Count)
92.4Pass@1 RateTIP
Evaluation results
| Method | Links | ||
|---|---|---|---|
| TIPModel=R1-7B2025.12 | 92.4 | 3,173 | |
| CRESTModel=R1-7B2025.12 | 92.4 | 2,661 | |
| DynasorModel=R1-7B2025.12 | 92 | 3,619 | |
| VanillaModel=R1-7B2025.12 | 91.6 | 4,020 | |
| VanillaBackbone=Qwen3-4B2026.03 | 91.6 | 5,021.6 | |
| SEALModel=R1-7B2025.12 | 91.2 | 3,335 | |
| TRiMSBackbone=Qwen3-4B2026.03 | 91 | 662.2 | |
| Soft-ThinkingModel=R1-7B2025.12 | 90 | 4,095 | |
| DynasorModel=R1-1.5B2025.12 | 89 | 3,267 | |
| CRESTModel=R1-1.5B2025.12 | 84.8 | 4,106 | |
| DSBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 84.6 | 4,010 | |
| InfoDensityBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 84.4 | 4,207 | |
| VanillaModel=R1-1.5B2025.12 | 84 | 5,497 | |
| GRPO-AccBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 84 | 4,456 | |
| PEARBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 83.8 | 3,744 | |
| TIPModel=R1-1.5B2025.12 | 83.4 | 4,414 | |
| GRPO-LPBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 82.8 | 4,734 | |
| OriginalBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 82.2 | 5,569 | |
| SEALModel=R1-1.5B2025.12 | 81.6 | 4,150 | |
| GRPO-AccBackbone=Qwen3-0.6B2026.03 | 75.8 | 5,574 | |
| OriginalBackbone=Qwen3-0.6B2026.03 | 73.2 | 5,745 | |
| PEARBackbone=Qwen3-0.6B2026.03 | 73.2 | 4,005 | |
| InfoDensityBackbone=Qwen3-0.6B2026.03 | 72.2 | 3,959 | |
| GRPO-LPBackbone=Qwen3-0.6B2026.03 | 71.4 | 4,604 | |
| DSBackbone=Qwen3-0.6B2026.03 | 70.2 | 4,187 | |
| Soft-ThinkingModel=R1-1.5B2025.12 | 66.8 | 9,401 |