Mathematics on MATH-500 (test)
80.2Score (%)Full
Evaluation Results
| Method | Links | |
|---|---|---|
| FullModel=Qwen3, Pruning Ratio=0%, Criterion=Full, Calibration=C42026.06 | 80.2 | |
| MSANModel=Qwen3, Pruning Ratio=25%, Criterion=MSAN, (b, α, β)=(1, 0, 2), Calibration=C42026.06 | 79.6 | |
| MANModel=Qwen3, Pruning Ratio=25%, Criterion=MAN, (b, α, β)=(1, 0, 1), Calibration=C42026.06 | 79.2 | |
| FullModel=ERNIE, Pruning Ratio=0%, Criterion=Full, Calibration=C42026.06 | 78 | |
| REAPModel=Qwen3, Pruning Ratio=25%, Criterion=REAP, (b, α, β)=(1, 1, 1), Calibration=C42026.06 | 77.8 | |
| MANModel=ERNIE, Pruning Ratio=25%, Criterion=MAN, (b, α, β)=(1, 0, 1), Calibration=C42026.06 | 58 | |
| MSANModel=ERNIE, Pruning Ratio=25%, Criterion=MSAN, (b, α, β)=(1, 0, 2), Calibration=C42026.06 | 56.4 | |
| FReDA-4BNumber of shots=4, Evaluation protocol=Best-of-N2026.06 | 53 | |
| REAPModel=ERNIE, Pruning Ratio=25%, Criterion=REAP, (b, α, β)=(1, 1, 1), Calibration=C42026.06 | 52.8 | |
| Qwen3-4B-BaseNumber of shots=42026.06 | 52.6 | |
| FReDA-4BNumber of shots=4, Evaluation protocol=Self-refine2026.06 | 51.98 | |
| BlockDiff-4BNumber of shots=42026.06 | 50.6 | |
| Qwen2.5-3B-BaseNumber of shots=42026.06 | 42.6 | |
| LLaDA-MoE-7B-A1B-BaseNumber of shots=42026.06 | 41.2 | |
| LLaDA-8B-BaseNumber of shots=42026.06 | 39.6 | |
| Dream-7B-BaseNumber of shots=42026.06 | 39.2 | |
| MoNEModel=ERNIE, Pruning Ratio=25%, Criterion=MoNE, Calibration=C42026.06 | 38.4 | |
| EANModel=ERNIE, Pruning Ratio=25%, Criterion=EAN, (b, α, β)=(0, 0, 1), Calibration=C42026.06 | 37 | |
| SEERModel=ERNIE, Pruning Ratio=25%, Criterion=SEER, (b, α, β)=(0, 1, 0), Calibration=C42026.06 | 36.8 | |
| FrequencyModel=ERNIE, Pruning Ratio=25%, Criterion=Frequency, (b, α, β)=(0, 0, 0), Calibration=C42026.06 | 31.6 | |
| FullModel=DeepSeek, Pruning Ratio=0%, Criterion=Full, Calibration=C42026.06 | 29.8 | |
| FullModel=OLMoE, Pruning Ratio=0%, Criterion=Full, Calibration=C42026.06 | 22.2 | |
| MoNEModel=Qwen3, Pruning Ratio=25%, Criterion=MoNE, Calibration=C42026.06 | 20.6 | |
| SEERModel=Qwen3, Pruning Ratio=25%, Criterion=SEER, (b, α, β)=(0, 1, 0), Calibration=C42026.06 | 20.2 | |
| FrequencyModel=Qwen3, Pruning Ratio=25%, Criterion=Frequency, (b, α, β)=(0, 0, 0), Calibration=C42026.06 | 19.6 | |
| EANModel=Qwen3, Pruning Ratio=25%, Criterion=EAN, (b, α, β)=(0, 0, 1), Calibration=C42026.06 | 19.4 | |
| DiffuLLaMA-7BNumber of shots=42026.06 | 11 | |
| MSANModel=DeepSeek, Pruning Ratio=25%, Criterion=MSAN, (b, α, β)=(1, 0, 2), Calibration=C42026.06 | 10.2 | |
| MSANModel=OLMoE, Pruning Ratio=25%, Criterion=MSAN, (b, α, β)=(1, 0, 2), Calibration=C42026.06 | 5.6 | |
| MANModel=OLMoE, Pruning Ratio=25%, Criterion=MAN, (b, α, β)=(1, 0, 1), Calibration=C42026.06 | 4.6 | |
| REAPModel=OLMoE, Pruning Ratio=25%, Criterion=REAP, (b, α, β)=(1, 1, 1), Calibration=C42026.06 | 3.6 | |
| MANModel=DeepSeek, Pruning Ratio=25%, Criterion=MAN, (b, α, β)=(1, 0, 1), Calibration=C42026.06 | 3.2 | |
| EANModel=DeepSeek, Pruning Ratio=25%, Criterion=EAN, (b, α, β)=(0, 0, 1), Calibration=C42026.06 | 2.8 | |
| REAPModel=DeepSeek, Pruning Ratio=25%, Criterion=REAP, (b, α, β)=(1, 1, 1), Calibration=C42026.06 | 2.8 | |
| FrequencyModel=OLMoE, Pruning Ratio=25%, Criterion=Frequency, (b, α, β)=(0, 0, 0), Calibration=C42026.06 | 2.4 | |
| MoNEModel=DeepSeek, Pruning Ratio=25%, Criterion=MoNE, Calibration=C42026.06 | 2.4 | |
| SEERModel=DeepSeek, Pruning Ratio=25%, Criterion=SEER, (b, α, β)=(0, 1, 0), Calibration=C42026.06 | 1.6 | |
| SEERModel=OLMoE, Pruning Ratio=25%, Criterion=SEER, (b, α, β)=(0, 1, 0), Calibration=C42026.06 | 1.2 | |
| EANModel=OLMoE, Pruning Ratio=25%, Criterion=EAN, (b, α, β)=(0, 0, 1), Calibration=C42026.06 | 1.2 | |
| FrequencyModel=DeepSeek, Pruning Ratio=25%, Criterion=Frequency, (b, α, β)=(0, 0, 0), Calibration=C42026.06 | 1.2 | |
| MoNEModel=OLMoE, Pruning Ratio=25%, Criterion=MoNE, Calibration=C42026.06 | 0.6 |