Mathematical Reasoning on MATH 500 (Acc, Tok #N, CR, Think #R)
96.5AccTA
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| TAModel Base Pair=Qwen3-4B-2507-Thinking/Qwen3-4B-2507-Instruct, Sampling Rounds=2x2026.01 | 96.5 | 4,671 | 70 | 100 | |
| MI-07Model Base Pair=Qwen3-4B-2507-Thinking/Qwen3-4B-2507-Instruct, Sampling Rounds=2x2026.01 | 96.4 | 4,617 | 69 | 100 | |
| RoutingModel Base Pair=Qwen3-4B-2507-Thinking/Qwen3-4B-2507-Instruct, Sampling Rounds=2x2026.01 | 96.4 | 5,435 | 82 | 53.8 | |
| DAMI-PrefModel Base Pair=Qwen3-4B-2507-Thinking/Qwen3-4B-2507-Instruct, Sampling Rounds=2x2026.01 | 96.4 | 4,315 | 65 | 64 | |
| DAMI-ConfModel Base Pair=Qwen3-4B-2507-Thinking/Qwen3-4B-2507-Instruct, Sampling Rounds=2x2026.01 | 96.2 | 2,954 | 44 | 34 | |
| ThinkingModel Base Pair=Qwen3-4B-2507-Thinking/Qwen3-4B-2507-Instruct, Sampling Rounds=2x2026.01 | 96.1 | 6,668 | 100 | 100 | |
| MI-05Model Base Pair=Qwen3-4B-2507-Thinking/Qwen3-4B-2507-Instruct, Sampling Rounds=2x2026.01 | 96.1 | 3,213 | 48 | 72 | |
| DAMI-PredModel Base Pair=Qwen3-4B-2507-Thinking/Qwen3-4B-2507-Instruct, Sampling Rounds=2x2026.01 | 95.6 | 1,886 | 28 | 0 | |
| DEERModel Base Pair=Qwen3-4B-2507-Thinking/Qwen3-4B-2507-Instruct, Sampling Rounds=2x2026.01 | 95.4 | 6,125 | 92 | 100 | |
| MI-03Model Base Pair=Qwen3-4B-2507-Thinking/Qwen3-4B-2507-Instruct, Sampling Rounds=2x2026.01 | 95.3 | 1,933 | 29 | 0 | |
| DAMI-PromptModel Base Pair=Qwen3-4B-2507-Thinking/Qwen3-4B-2507-Instruct, Sampling Rounds=2x2026.01 | 95 | 2,832 | 42 | 27 | |
| TIESModel Base Pair=Qwen3-4B-2507-Thinking/Qwen3-4B-2507-Instruct, Sampling Rounds=2x2026.01 | 94.8 | 1,899 | 28 | 0 | |
| InstructModel Base Pair=Qwen3-4B-2507-Thinking/Qwen3-4B-2507-Instruct, Sampling Rounds=2x2026.01 | 93.1 | 1,741 | 26.1 | 0 | |
| DAMI-PrefModel Base Pair=DeepSeek-R1-Distill-Qwen-7B/Qwen2.5-Math-7B, Sampling Rounds=2x2026.01 | 89.4 | 3,952 | 67 | 75 | |
| DAMI-ConfModel Base Pair=DeepSeek-R1-Distill-Qwen-7B/Qwen2.5-Math-7B, Sampling Rounds=2x2026.01 | 89 | 3,526 | 60 | 62 | |
| DEERModel Base Pair=DeepSeek-R1-Distill-Qwen-7B/Qwen2.5-Math-7B, Sampling Rounds=2x2026.01 | 88.8 | 3,140 | 53 | 100 | |
| TAModel Base Pair=DeepSeek-R1-Distill-Qwen-7B/Qwen2.5-Math-7B, Sampling Rounds=2x2026.01 | 88.8 | 2,670 | 45 | 97 | |
| MI-07Model Base Pair=DeepSeek-R1-Distill-Qwen-7B/Qwen2.5-Math-7B, Sampling Rounds=2x2026.01 | 88.2 | 3,243 | 55 | 95 | |
| ThinkingModel Base Pair=DeepSeek-R1-Distill-Qwen-7B/Qwen2.5-Math-7B, Sampling Rounds=2x2026.01 | 86 | 5,881 | 100 | 100 | |
| MI-05Model Base Pair=DeepSeek-R1-Distill-Qwen-7B/Qwen2.5-Math-7B, Sampling Rounds=2x2026.01 | 83.6 | 2,338 | 40 | 13 | |
| TIESModel Base Pair=DeepSeek-R1-Distill-Qwen-7B/Qwen2.5-Math-7B, Sampling Rounds=2x2026.01 | 80.2 | 827 | 14 | 0 | |
| DAMI-PredModel Base Pair=DeepSeek-R1-Distill-Qwen-7B/Qwen2.5-Math-7B, Sampling Rounds=2x2026.01 | 79.6 | 1,823 | 31 | 0 | |
| MI-03Model Base Pair=DeepSeek-R1-Distill-Qwen-7B/Qwen2.5-Math-7B, Sampling Rounds=2x2026.01 | 76.8 | 1,501 | 26 | 0 | |
| DAMI-PromptModel Base Pair=DeepSeek-R1-Distill-Qwen-7B/Qwen2.5-Math-7B, Sampling Rounds=2x2026.01 | 76.2 | 1,657 | 28 | 0 | |
| RoutingModel Base Pair=DeepSeek-R1-Distill-Qwen-7B/Qwen2.5-Math-7B, Sampling Rounds=2x2026.01 | 63.6 | 1,303 | 22 | 4.4 | |
| InstructModel Base Pair=DeepSeek-R1-Distill-Qwen-7B/Qwen2.5-Math-7B, Sampling Rounds=2x2026.01 | 52 | 1,268 | — | 0 | |
| BRIDGEBackbone=Qwen 2.5-3B2026.02 | 38.2 | 322 | — | — | |
| Base ModelBackbone=Qwen 2.5-3B2026.02 | 36.4 | 371 | — | — | |
| Instruct ModelBackbone=Qwen 2.5-3B2026.02 | 34.2 | 440 | — | — | |
| SuperRLBackbone=Qwen 2.5-3B2026.02 | 33.2 | 355 | — | — | |
| Std-CoT KDBackbone=Qwen 2.5-3B2026.02 | 26.6 | 367 | — | — | |
| Instruct ModelBackbone=Llama 3.2-3B2026.02 | 19.4 | 181 | — | — | |
| BRIDGEBackbone=Llama 3.2-3B2026.02 | 16 | 107 | — | — | |
| Mix Distill.Backbone=Qwen 2.5-3B2026.02 | 15.4 | 280 | — | — | |
| SuperRLBackbone=Llama 3.2-3B2026.02 | 8.6 | 245 | — | — | |
| Short-CoTBackbone=Qwen 2.5-3B2026.02 | 8.5 | 220 | — | — | |
| Std-CoT KDBackbone=Llama 3.2-3B2026.02 | 7.2 | 185 | — | — | |
| Mix Distill.Backbone=Llama 3.2-3B2026.02 | 6.4 | 149 | — | — | |
| Short-CoTBackbone=Llama 3.2-3B2026.02 | 6.2 | 185 | — | — | |
| Base ModelBackbone=Llama 3.2-3B2026.02 | 5.33 | 155 | — | — |