Mathematical Reasoning on MATH L5
0.859Accuracys1
Evaluation results
| Method | Links | ||
|---|---|---|---|
| s1Base Model=Qwen3-4B, SFT Setting=1K, Data Selection Mode=offline2026.01 | 0.859 | — | |
| RandomBase Model=Qwen3-4B, SFT Setting=1K2026.01 | 0.853 | — | |
| LIMOBase Model=Qwen3-4B, SFT Setting=1K, Data Selection Mode=offline2026.01 | 0.853 | — | |
| Skill-centric distillation frameworkBase Model=Qwen3-4B, SFT Setting=1K2026.01 | 0.853 | — | |
| s1Base Model=Qwen3-4B, SFT Setting=1K, Data Selection Mode=online2026.01 | 0.849 | — | |
| BaseBase Model=Qwen3-4B, SFT Setting=1K2026.01 | 0.843 | — | |
| Thought-ICSModel=OSS-20B2026.02 | 0.75 | — | |
| Qwen3-4BTotal Parameters=4B, Active Parameters=4B, Trained Tokens=36T2025.11 | 0.7362 | — | |
| Thought-ICSModel=OSS-120B2026.02 | 0.73 | — | |
| Thought-ICSModel=Qwen-32B2026.02 | 0.69 | — | |
| ARMOR-MADAgents=gpt-4o-mini, deepseek-v3, qwen-plus2026.06 | 0.655 | — | |
| Hetero-MADAgents=gpt-4o-mini, deepseek-v3, qwen-plus2026.06 | 0.635 | — | |
| gemma-3-27b-itType / λ=Base IT2025.10 | 0.6314 | — | |
| gemma3-27b-slerpType / λ=0.92025.10 | 0.6314 | — | |
| LFM2-8B-A1BTotal Parameters=8.3B, Active Parameters=1.5B, Trained Tokens=13T2025.11 | 0.6238 | — | |
| gemma3-27b-slerpType / λ=0.82025.10 | 0.6216 | — | |
| Thought-ICSModel=LLaMA-70B2026.02 | 0.62 | — | |
| MoA2026.06 | 0.605 | — | |
| Thought-ICS-AModel=OSS-20B2026.02 | 0.59 | — | |
| D-MAD2026.06 | 0.59 | — | |
| gemma3-27b-slerpType / λ=0.72025.10 | 0.5861 | — | |
| Thought-ICSModel=Qwen-7B2026.02 | 0.58 | — | |
| Thought-ICSModel=Qwen-14B2026.02 | 0.58 | — | |
| Thought-ICS-AModel=OSS-120B2026.02 | 0.57 | — | |
| gemma3-27b-slerpType / λ=0.62025.10 | 0.565 | — | |
| CoVeModel=Qwen-32B2026.02 | 0.56 | — | |
| Thought-ICS-SModel=Qwen-32B2026.02 | 0.56 | — | |
| Thought-ICS-AModel=Qwen-32B2026.02 | 0.56 | — | |
| gemma3-12b-linearType / λ=0.92025.10 | 0.5589 | — | |
| gemma-3-12b-itType / λ=Base IT2025.10 | 0.5582 | — | |
| gemma3-12b-slerpType / λ=0.92025.10 | 0.5506 | — | |
| Token-ICSModel=Qwen-32B2026.02 | 0.55 | — | |
| LFM2-2.6BTotal Parameters=2.6B, Active Parameters=2.6B, Trained Tokens=11T2025.11 | 0.5438 | — | |
| Sigmoid capability boundariesFLOPs budget=10^24, Boundary estimation method=no-split 0.98-quantile sigmoid2026.02 | 0.539 | — | |
| Homo-MADAgents=gpt-4o-mini2026.06 | 0.535 | — | |
| CoVeModel=OSS-20B2026.02 | 0.53 | — | |
| MV@10Model=Qwen-14B2026.02 | 0.53 | — | |
| gemma3-12b-slerpType / λ=0.82025.10 | 0.5287 | — | |
| gemma3-27b-slerpType / λ=0.52025.10 | 0.5264 | — | |
| Gemma-3-4BTotal Parameters=4B, Active Parameters=4B, Trained Tokens=4T2025.11 | 0.5218 | — | |
| Self-RefineModel=OSS-20B2026.02 | 0.52 | — | |
| Thought-ICS-SModel=OSS-20B2026.02 | 0.52 | — | |
| Thought-ICS-SModel=OSS-120B2026.02 | 0.52 | — | |
| MV@10Model=Qwen-32B2026.02 | 0.52 | — | |
| SmolLM3-3BTotal Parameters=3.1B, Active Parameters=3.1B, Trained Tokens=11T2025.11 | 0.5193 | — | |
| gemma3-12b-linearType / λ=0.82025.10 | 0.5189 | — | |
| gemma3-12b-dare_tiesType / λ=0.82025.10 | 0.5159 | — | |
| gemma3-12b-dare_tiesType / λ=0.92025.10 | 0.506 | — | |
| Hetero-Vote-onlyAgents=gpt-4o-mini, deepseek-v3, qwen-plus2026.06 | 0.505 | — | |
| Self-ConsistencyAgents=gpt-4o-mini2026.06 | 0.5 | — | |
| gemma3-12b-dare_tiesType / λ=0.72025.10 | 0.4917 | — | |
| Token-ICSModel=LLaMA-70B2026.02 | 0.49 | — | |
| gemma3-27b-slerpType / λ=0.42025.10 | 0.4887 | — | |
| gemma3-12b-linearType / λ=0.72025.10 | 0.4872 | — | |
| gemma3-12b-slerpType / λ=0.72025.10 | 0.4849 | — | |
| MV@10Model=LLaMA-70B2026.02 | 0.48 | — | |
| gemma3-12b-linearType / λ=0.62025.10 | 0.4721 | — | |
| Self-RefineModel=Qwen-32B2026.02 | 0.47 | — | |
| CoVeModel=OSS-120B2026.02 | 0.47 | — | |
| MV@10Model=Qwen-7B2026.02 | 0.47 | — | |
| Token-ICSModel=Qwen-14B2026.02 | 0.47 | — | |
| Token-ICSModel=OSS-120B2026.02 | 0.47 | — | |
| gemma3-12b-slerpType / λ=0.62025.10 | 0.4675 | — | |
| Token-ICSModel=OSS-20B2026.02 | 0.46 | — | |
| CoTModel=OSS-20B2026.02 | 0.45 | — | |
| InitModel=OSS-20B2026.02 | 0.45 | — | |
| MV@10Model=OSS-20B2026.02 | 0.45 | — | |
| gemma3-12b-dare_tiesType / λ=0.62025.10 | 0.4486 | — | |
| Self-RefineModel=OSS-120B2026.02 | 0.44 | — | |
| gemma3-27b-slerpType / λ=0.32025.10 | 0.4381 | — | |
| gemma3-12b-linearType / λ=0.52025.10 | 0.4328 | — | |
| Thought-ICS-SModel=Qwen-14B2026.02 | 0.43 | — | |
| Thought-ICS-AModel=Qwen-14B2026.02 | 0.43 | — | |
| Single-CoTAgents=gpt-4o-mini2026.06 | 0.43 | — | |
| gemma3-12b-slerpType / λ=0.52025.10 | 0.4298 | — | |
| CoTModel=OSS-120B2026.02 | 0.42 | — | |
| InitModel=OSS-120B2026.02 | 0.42 | — | |
| gemma3-12b-dare_tiesType / λ=0.52025.10 | 0.4199 | — | |
| CoTModel=Qwen-32B2026.02 | 0.41 | — | |
| InitModel=Qwen-32B2026.02 | 0.41 | — | |
| Thought-ICS-SModel=Qwen-7B2026.02 | 0.4 | — | |
| Thought-ICS-AModel=Qwen-7B2026.02 | 0.4 | — | |
| CoTModel=Qwen-14B2026.02 | 0.4 | — | |
| Self-RefineModel=Qwen-14B2026.02 | 0.4 | — | |
| InitModel=Qwen-14B2026.02 | 0.4 | — | |
| gemma3-12b-linearType / λ=0.42025.10 | 0.3912 | — | |
| CoTModel=LLaMA-70B2026.02 | 0.39 | — | |
| InitModel=LLaMA-70B2026.02 | 0.39 | — | |
| gemma3-12b-slerpType / λ=0.42025.10 | 0.3875 | — | |
| Self-RefineModel=Qwen-7B2026.02 | 0.38 | — | |
| Thought-ICS-SModel=LLaMA-70B2026.02 | 0.38 | — | |
| Thought-ICS-AModel=LLaMA-70B2026.02 | 0.38 | — | |
| Token-ICSModel=Qwen-7B2026.02 | 0.38 | — | |
| Thought-ICSModel=LLaMA-8B2026.02 | 0.38 | — | |
| gemma3-12b-dare_tiesType / λ=0.42025.10 | 0.3784 | — | |
| gemma3-27b-linearType / λ=0.22025.10 | 0.3739 | — | |
| CoVeModel=Qwen-14B2026.02 | 0.37 | — | |
| CoVeModel=LLaMA-70B2026.02 | 0.37 | — | |
| gemma3-27b-slerpType / λ=0.22025.10 | 0.3671 | — | |
| Granite-4.0-HTotal Parameters=7B, Active Parameters=1B, Trained Tokens=15T2025.11 | 0.3611 | — |