Mathematical Reasoning on MATH500 (Pass@1, FS, N/T, D/T, Repro)
97.9Pass@1 AccuracyRED
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| REDBackbone=Qwen3-32B-Thinking2026.04 | 97.9 | 1.9 | 2.5 | 1.9 | 0.016 | |
| S-GRROBackbone=Qwen3-32B-Thinking2026.04 | 97.3 | 3.7 | 4.2 | 3.1 | 0.044 | |
| RL + LPBackbone=Qwen3-32B-Thinking2026.04 | 97.2 | 3.5 | 4.1 | 2.8 | 0.043 | |
| VanillaBackbone=Qwen3-32B-Thinking2026.04 | 96.8 | 4.1 | 4.4 | 3.2 | 0.049 | |
| DEERBackbone=Qwen3-32B-Thinking2026.04 | 96.4 | 3.8 | 4.3 | 3.2 | 0.05 | |
| REDBackbone=R1-Llama-70B2026.04 | 96.2 | 2.3 | 2.8 | 2.2 | 0.024 | |
| REDBackbone=R1-Qwen-32B2026.04 | 95.2 | 1.9 | 2.8 | 2.2 | 0.024 | |
| S-GRROBackbone=R1-Llama-70B2026.04 | 95.1 | 4.7 | 4.6 | 3.7 | 0.069 | |
| S-GRROBackbone=R1-Qwen-32B2026.04 | 94.2 | 3.8 | 4.4 | 3.4 | 0.053 | |
| VanillaBackbone=R1-Llama-70B2026.04 | 94.1 | 5.2 | 5 | 3.8 | 0.075 | |
| RL + LPBackbone=R1-Llama-70B2026.04 | 93.8 | 4.6 | 4.5 | 3.5 | 0.066 | |
| VanillaBackbone=R1-Qwen-32B2026.04 | 93.3 | 4.2 | 4.7 | 3.5 | 0.057 | |
| RL + LPBackbone=R1-Qwen-32B2026.04 | 93.1 | 3.7 | 4.4 | 3.1 | 0.051 | |
| DEERBackbone=R1-Llama-70B2026.04 | 92.3 | 4.7 | 4.7 | 3.9 | 0.077 | |
| DEERBackbone=R1-Qwen-32B2026.04 | 91.8 | 3.9 | 4.6 | 3.5 | 0.058 |