Mathematical Reasoning on MATH 500 (∆Acc, CR)
100Completion RateVanilla
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| VanillaAccess=—, Base Model=Qwen3-14B (DEER) / Qwen3-8B-Think (ours)2026.04 | 100 | — | |
| VanillaAccess=—, Base Model=QwQ-32B (DEER) / Qwen3-32B-Think (ours)2026.04 | 100 | — | |
| Dynasor-CoTAccess=black-box, Base Model=Qwen3-14B (DEER) / Qwen3-8B-Think (ours)2026.04 | 72 | 0.2 | |
| DEER-PRoAccess=semi-white, Base Model=QwQ-32B (DEER) / Qwen3-32B-Think (ours)2026.04 | 72 | 1 | |
| DEERAccess=semi-white, Base Model=QwQ-32B (DEER) / Qwen3-32B-Think (ours)2026.04 | 69 | 0.5 | |
| TCCAccess=black-box, Base Model=Qwen3-14B (DEER) / Qwen3-8B-Think (ours)2026.04 | 61 | 0.6 | |
| DEER-PRoAccess=semi-white, Base Model=Qwen3-14B (DEER) / Qwen3-8B-Think (ours)2026.04 | 45 | 2.8 | |
| DEERAccess=semi-white, Base Model=Qwen3-14B (DEER) / Qwen3-8B-Think (ours)2026.04 | 41 | 0.5 | |
| BAEEAccess=black-box, Base Model=Qwen3-14B (DEER) / Qwen3-8B-Think (ours)2026.04 | 32 | 4.4 | |
| BAEEAccess=black-box, Base Model=QwQ-32B (DEER) / Qwen3-32B-Think (ours)2026.04 | 28 | 3 | |
| NoThinkingAccess=black-box, Base Model=Qwen3-14B (DEER) / Qwen3-8B-Think (ours)2026.04 | 27 | 5.6 |