Math Reasoning on AMC (Pass@1, Pass@16)
88.6Pass@1NeuReasoner
Evaluation results
| Method | Links | ||
|---|---|---|---|
| NeuReasonerModel Size=Qwen3-32B, Condition=our cognitive scaffold, thinking off2026.06 | 88.6 | — | |
| Thinking onRLModel Size=Qwen3-8B, Condition=native RL-trained reasoning mode2026.06 | 87.6 | — | |
| Thinking onRLModel Size=Qwen3-32B, Condition=native RL-trained reasoning mode2026.06 | 86.3 | — | |
| GSPO + NSRModel=Qwen3-30B-A3B-Base, Setting=MoE + 20k, Zero-shot=true, k=32, Temperature=1.02026.05 | 85.09 | 94.37 | |
| GSPOModel=Qwen3-30B-A3B-Base, Setting=MoE + 20k, Zero-shot=true, k=32, Temperature=1.02026.05 | 81.66 | 92.01 | |
| DAPO + NSRModel=Qwen2.5-Math-7B, Setting=Dense + 8k, Zero-shot=true, k=32, Temperature=1.02026.05 | 76.74 | 89.36 | |
| EAPOModel=Qwen-3-8B-Base2026.06 | 76.54 | 91.51 | |
| DAPO + NSRModel=Qwen3-8B-Base, Setting=Dense + 30k, Zero-shot=true, k=32, Temperature=1.02026.05 | 74.92 | 91.28 | |
| EAPOModel=Qwen-2.5-Math-7B2026.06 | 72.81 | 89.74 | |
| Trajectory ReplayModel=Qwen-2.5-Math-7B2026.06 | 72.74 | 89.93 | |
| EAPO w/o sISModel=Qwen-2.5-Math-7B2026.06 | 72.66 | 88.27 | |
| MOPDModel=Qwen-3-8B-Base2026.06 | 72.03 | 89.83 | |
| DAPOModel=Qwen-3-8B-Base2026.06 | 71.88 | 89.87 | |
| DAPOModel=Qwen3-8B-Base, Setting=Dense + 30k, Zero-shot=true, k=32, Temperature=1.02026.05 | 71.42 | 88.52 | |
| OPDModel=Qwen-3-8B-Base2026.06 | 71.16 | 88.17 | |
| Thinking offModel Size=Qwen3-32B, Condition=vanilla chain-of-thought, no scaffold2026.06 | 69.5 | — | |
| DAPOModel=Qwen2.5-Math-7B, Setting=Dense + 8k, Zero-shot=true, k=32, Temperature=1.02026.05 | 69.28 | 88.24 | |
| DAPOModel=Qwen-2.5-Math-7B2026.06 | 68.86 | 89.59 | |
| OPDModel=Qwen-2.5-Math-7B2026.06 | 68.59 | 87.75 | |
| MOPDModel=Qwen-2.5-Math-7B2026.06 | 68.56 | 87.67 | |
| GRPOModel=Qwen-3-8B-Base2026.06 | 67.73 | 86.89 | |
| GRPOModel=Qwen-2.5-Math-7B2026.06 | 67.39 | 81.24 | |
| Thinking offModel Size=Qwen3-8B, Condition=vanilla chain-of-thought, no scaffold2026.06 | 59.8 | — | |
| NeuReasonerModel Size=Qwen3-8B, Condition=our cognitive scaffold, thinking off2026.06 | 50.3 | — |