Mathematical Reasoning on AIME 24 (Accuracy, Tokens, Delta %)
76.7AccuracyCoD
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| CoDReasoner=Qwen3-8B2026.06 | 76.7 | 12,693 | 14.7 | |
| VanillaReasoner=Qwen3-8B2026.06 | 76 | 14,880 | — | |
| ACTSπSFTReasoner=Qwen3-8B2026.06 | 74 | 12,727 | 14.5 | |
| ACTSReasoner=Qwen3-8B2026.06 | 73.3 | 11,198 | 24.7 | |
| DEERReasoner=Qwen3-8B2026.06 | 70 | 11,873 | 20.2 | |
| ACTSReasoner=DeepSeek-R1-Distill-Qwen-7B2026.06 | 60 | 8,200 | 37.1 | |
| ACTSπSFTReasoner=DeepSeek-R1-Distill-Qwen-7B2026.06 | 56.7 | 8,693 | 33.3 | |
| VanillaReasoner=DeepSeek-R1-Distill-Qwen-7B2026.06 | 51.3 | 13,031 | — | |
| BudgetGuidanceReasoner=DeepSeek-R1-Distill-Qwen-7B2026.06 | 50.7 | 11,694 | 10.3 | |
| DEERReasoner=DeepSeek-R1-Distill-Qwen-7B2026.06 | 50 | 9,411 | 27.8 | |
| CoDReasoner=DeepSeek-R1-Distill-Qwen-7B2026.06 | 46 | 11,427 | 12.3 | |
| BudgetGuidanceReasoner=Qwen3-8B2026.06 | 44 | 8,395 | 43.6 | |
| ACTSReasoner=DeepSeek-R1-Distill-Qwen-1.5B2026.06 | 36.7 | 10,417 | 37.9 | |
| ACTSπSFTReasoner=DeepSeek-R1-Distill-Qwen-1.5B2026.06 | 35.3 | 10,977 | 34.6 | |
| DEERReasoner=DeepSeek-R1-Distill-Qwen-1.5B2026.06 | 30.7 | 8,392 | 50 | |
| NoThinkReasoner=Qwen3-8B2026.06 | 30 | 6,475 | 56.5 | |
| VanillaReasoner=DeepSeek-R1-Distill-Qwen-1.5B2026.06 | 28 | 16,776 | — | |
| CoDReasoner=DeepSeek-R1-Distill-Qwen-1.5B2026.06 | 26.7 | 12,891 | 23.2 | |
| NoThinkReasoner=DeepSeek-R1-Distill-Qwen-7B2026.06 | 25.3 | 3,944 | 69.7 | |
| NoThinkReasoner=DeepSeek-R1-Distill-Qwen-1.5B2026.06 | 14 | 5,679 | 66.2 |