Mathematical Reasoning on OlympiadBench (Acc, #Tok, ∆%)
71.4AccuracyDEER
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| DEERReasoner=Qwen3-8B2026.06 | 71.4 | 9,408 | 17.2 | |
| VanillaReasoner=Qwen3-8B2026.06 | 67.7 | 11,368 | — | |
| ACTSπSFTReasoner=Qwen3-8B2026.06 | 67.4 | 8,799 | 22.6 | |
| ACTSReasoner=Qwen3-8B2026.06 | 66.5 | 8,322 | 26.8 | |
| CoDReasoner=Qwen3-8B2026.06 | 66.1 | 8,684 | 23.6 | |
| VanillaReasoner=DeepSeek-R1-Distill-Qwen-7B2026.06 | 59.6 | 8,532 | — | |
| BudgetGuidanceReasoner=Qwen3-8B2026.06 | 59.6 | 4,033 | 64.5 | |
| ACTSReasoner=DeepSeek-R1-Distill-Qwen-7B2026.06 | 58.1 | 4,181 | 51 | |
| BudgetGuidanceReasoner=DeepSeek-R1-Distill-Qwen-7B2026.06 | 56.7 | 6,890 | 19.2 | |
| ACTSπSFTReasoner=DeepSeek-R1-Distill-Qwen-7B2026.06 | 56.4 | 5,206 | 39 | |
| NoThinkReasoner=Qwen3-8B2026.06 | 55 | 2,837 | 75 | |
| DEERReasoner=DeepSeek-R1-Distill-Qwen-7B2026.06 | 51 | 4,994 | 41.5 | |
| ACTSReasoner=DeepSeek-R1-Distill-Qwen-1.5B2026.06 | 45.9 | 4,460 | 64.1 | |
| CoDReasoner=DeepSeek-R1-Distill-Qwen-7B2026.06 | 45.8 | 5,716 | 33 | |
| ACTSπSFTReasoner=DeepSeek-R1-Distill-Qwen-1.5B2026.06 | 44.9 | 4,561 | 63.3 | |
| NoThinkReasoner=DeepSeek-R1-Distill-Qwen-7B2026.06 | 44.7 | 1,645 | 80.7 | |
| VanillaReasoner=DeepSeek-R1-Distill-Qwen-1.5B2026.06 | 42.5 | 12,418 | — | |
| NoThinkReasoner=DeepSeek-R1-Distill-Qwen-1.5B2026.06 | 38.1 | 2,162 | 82.6 | |
| CoDReasoner=DeepSeek-R1-Distill-Qwen-1.5B2026.06 | 33.9 | 8,499 | 31.6 | |
| DEERReasoner=DeepSeek-R1-Distill-Qwen-1.5B2026.06 | 33.5 | 4,814 | 61.2 |