Mathematical Reasoning on AIME 25 (Accuracy, Token consumption)
80AccuracyCoT-SC
Evaluation results
| Method | Links | ||
|---|---|---|---|
| CoT-SCBackbone=QwQ-32B2026.03 | 80 | 0.85 | |
| MAD-M2 (O)Backbone=QwQ-32B, Masking Strategy=Objective2026.03 | 76.7 | 0.9 | |
| MADBackbone=QwQ-32B2026.03 | 73.3 | 1 | |
| MAD-M2 (S)Backbone=QwQ-32B, Masking Strategy=Subjective2026.03 | 73.3 | 1.25 | |
| CoTBackbone=QwQ-32B2026.03 | 56.7 | 0.14 | |
| MAD-M2 (O)Backbone=Qwen2.5-Math-7B-Instruct, Masking Strategy=Objective2026.03 | 13.3 | 0.62 | |
| CoT-SCBackbone=Qwen2.5-7B-Instruct2026.03 | 10 | 0.45 | |
| CoTBackbone=Qwen2.5-Math-7B-Instruct2026.03 | 10 | 0.08 | |
| CoT-SCBackbone=Qwen2.5-Math-7B-Instruct2026.03 | 10 | 0.48 | |
| MADBackbone=Qwen2.5-7B-Instruct2026.03 | 6.7 | 1 | |
| MAD-M2 (O)Backbone=Qwen2.5-7B-Instruct, Masking Strategy=Objective2026.03 | 6.7 | 0.65 | |
| MADBackbone=Qwen2.5-Math-7B-Instruct2026.03 | 6.7 | 1 | |
| MAD-M2 (S)Backbone=Qwen2.5-Math-7B-Instruct, Masking Strategy=Subjective2026.03 | 6.7 | 1.37 | |
| CoTBackbone=Qwen2.5-7B-Instruct2026.03 | 3.3 | 0.08 | |
| MAD-M2 (S)Backbone=Qwen2.5-7B-Instruct, Masking Strategy=Subjective2026.03 | 3.3 | 1.21 | |
| CoTBackbone=DeepSeek-Math-7B-Instruct2026.03 | 0 | 0.09 | |
| CoT-SCBackbone=DeepSeek-Math-7B-Instruct2026.03 | 0 | 0.46 | |
| MADBackbone=DeepSeek-Math-7B-Instruct2026.03 | 0 | 1 | |
| MAD-M2 (S)Backbone=DeepSeek-Math-7B-Instruct, Masking Strategy=Subjective2026.03 | 0 | 1.3 | |
| MAD-M2 (O)Backbone=DeepSeek-Math-7B-Instruct, Masking Strategy=Objective2026.03 | 0 | 0.67 |