Mathematical Reasoning on AMC 2023 (avg@64, pass@64)
94.9Avg@64ASPO-Math-30B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ASPO-Math-30BBackbone=Qwen3-30B-A3B2025.10 | 94.9 | 97.5 | |
| GRPOBackbone=Qwen3-30B-A3B2025.10 | 91.9 | 97.5 | |
| ASPO-Math-8BBackbone=Qwen3-8B2025.10 | 89.1 | 97.5 | |
| ASPO-Math-4BBackbone=Qwen3-4B2025.10 | 87.4 | 97.5 | |
| ASPO-Math-1.5BBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.10 | 87.2 | 95 | |
| Nemotron-1.5BBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.10 | 86.1 | 97.5 | |
| GRPOBackbone=Qwen3-4B2025.10 | 84.3 | 97.5 | |
| GRPOBackbone=Qwen3-8B2025.10 | 84.1 | 95 | |
| DeepScaleR-1.5BBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.10 | 81.3 | 100 | |
| GRPOBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.10 | 80.3 | 97.5 | |
| Qwen3-30B-A3BBackbone=Qwen3-30B-A3B2025.10 | 74.3 | 100 | |
| DeepSeek-R1-1.5BBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.10 | 70.7 | 100 | |
| Qwen3-8BBackbone=Qwen3-8B2025.10 | 68.9 | 97.5 | |
| Qwen3-4BBackbone=Qwen3-4B2025.10 | 67.7 | 95 |