Mathematics on OlympiadBench (pass@1 %)
78.72Pass@1 AccuracyCOSE
Evaluation Results
| Method | Links | |
|---|---|---|
| COSEBackbone=Qwen3-4B2026.05 | 78.72 | |
| COSEBackbone=Llama-3.2-3B-Instruct2026.05 | 72 | |
| MAEBackbone=Qwen3-4B2026.05 | 69.4 | |
| R-ZeroBackbone=Qwen3-4B2026.05 | 68.9 | |
| AZRBackbone=Qwen3-4B2026.05 | 68.6 | |
| BaseBackbone=Qwen3-4B2026.05 | 66.8 | |
| INSIGHTModel=QWEN3-4B2026.03 | 57 | |
| MoPPSModel=QWEN3-4B2026.03 | 56.45 | |
| RANDOMModel=QWEN3-4B2026.03 | 56.24 | |
| INVERSE-EVIDENCEModel=QWEN3-4B2026.03 | 56.2 | |
| DSModel=R1-DISTILL-QWEN-7B2026.03 | 53.38 | |
| INSIGHTModel=R1-DISTILL-QWEN-7B2026.03 | 52.5 | |
| EXPECTED-DIFFICULTYModel=R1-DISTILL-QWEN-7B2026.03 | 52.25 | |
| INVERSE-EVIDENCEModel=R1-DISTILL-QWEN-7B2026.03 | 52.07 | |
| MoPPSModel=R1-DISTILL-QWEN-7B2026.03 | 52.01 | |
| RANDOMModel=R1-DISTILL-QWEN-7B2026.03 | 51.8 | |
| Scaf-GRPOBase Model=DeepSeek-R1-Distill-Qwen-1.5B2025.10 | 50.7 | |
| Vanilla GRPOBase Model=DeepSeek-R1-Distill-Qwen-1.5B2025.10 | 50.2 | |
| MAEBackbone=Llama-3.2-3B-Instruct2026.05 | 47.6 | |
| R-ZeroBackbone=Llama-3.2-3B-Instruct2026.05 | 46.3 | |
| DeepSeek-R1-Distill-Qwen-1.5BBase Model=DeepSeek-R1-Distill-Qwen-1.5B2025.10 | 45.8 | |
| AZRBackbone=Llama-3.2-3B-Instruct2026.05 | 44.2 | |
| Scaf-GRPOBase Model=Qwen2.5-Math-7B2025.10 | 43.3 | |
| LUFFYBase Model=Qwen2.5-Math-7B2025.10 | 41.7 | |
| Vanilla GRPOBase Model=Qwen2.5-Math-7B2025.10 | 41.3 | |
| Oat-ZeroBase Model=Qwen2.5-Math-7B2025.10 | 41 | |
| Scaf-GRPOBase Model=Qwen2.5-7B2025.10 | 40.8 | |
| Vanilla GRPOBase Model=Qwen2.5-7B2025.10 | 40.4 | |
| BaseBackbone=Llama-3.2-3B-Instruct2026.05 | 38.1 | |
| SimpleRL-ZeroBase Model=Qwen2.5-Math-7B2025.10 | 37.2 | |
| Scaf-GRPOBase Model=Qwen2.5-Math-1.5B2025.10 | 36.6 | |
| Vanilla GRPOBase Model=Qwen2.5-Math-1.5B2025.10 | 34.8 | |
| INVERSE-EVIDENCEModel=QWEN3-0.6B2026.03 | 34.8 | |
| INSIGHTModel=QWEN3-0.6B2026.03 | 34.8 | |
| Qwen2.5-7BBase Model=Qwen2.5-7B2025.10 | 34.4 | |
| MoPPSModel=QWEN3-0.6B2026.03 | 34.25 | |
| RANDOMModel=QWEN3-0.6B2026.03 | 34.2 | |
| Qwen2.5-Math-1.5BBase Model=Qwen2.5-Math-1.5B2025.10 | 20.6 | |
| Scaf-GRPOBase Model=Llama-3.2-3B-Instruct2025.10 | 20.3 | |
| IPPO (MAPoRL)Backbone=LaMAS, Scale=Trio2025.04 | 19.05 | |
| MARFTBackbone=LaMAS, Scale=Trio2025.04 | 18.72 | |
| MARFTBackbone=LaMAS, Scale=Quad2025.04 | 18.55 | |
| MARFTBackbone=LaMAS, Scale=Duo2025.04 | 18.43 | |
| IPPO (MAPoRL)Backbone=LaMAS, Scale=Quad2025.04 | 18.37 | |
| IPPO (MAPoRL)Backbone=LaMAS, Scale=Duo2025.04 | 18.31 | |
| Vanilla GRPOBase Model=Llama-3.2-3B-Instruct2025.10 | 18.3 | |
| Qwen2.5-Math-7BBase Model=Qwen2.5-Math-7B2025.10 | 18.2 | |
| VanillaBackbone=LaMAS, Scale=Trio2025.04 | 17.6 | |
| VanillaBackbone=LaMAS, Scale=Quad2025.04 | 16.6 | |
| VanillaBackbone=LaMAS, Scale=Duo2025.04 | 16.08 | |
| Llama-3.2-3B-InstructBase Model=Llama-3.2-3B-Instruct2025.10 | 12.6 |