Mathematical Reasoning on MATH 500 (Accuracy and Length Metrics)
93.8Accuracy (%)SPOT-stage2
Evaluation results
| Method | Links | ||
|---|---|---|---|
| SPOT-stage2Backbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 93.8 | 2,154 | |
| Step EntropyBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 92.8 | 2,649 | |
| ConciseHintBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 92.62 | 2,447 | |
| DEERBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 92.6 | 2,486 | |
| VanillaBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 92.4 | 3,775 | |
| CCoTBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 92.26 | 3,133 | |
| SPOT-stage1Backbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 92.24 | 1,626 | |
| LightThinkerBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 91.42 | 3,064 | |
| L1-MaxBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 89.4 | 2,580 | |
| Latent-SFTBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 79.8 | 787 | |
| CODIBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 76.6 | 624 | |
| COCONUTBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 69.88 | 494 |