Mathematical Reasoning on AIME 2025 (avg@64, pass@64)
50.3Avg@64ASPO-Math-30B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ASPO-Math-30BBackbone=Qwen3-30B-A3B2025.10 | 50.3 | 80 | |
| GRPOBackbone=Qwen3-30B-A3B2025.10 | 43.1 | 73.3 | |
| ASPO-Math-4BBackbone=Qwen3-4B2025.10 | 40.9 | 70 | |
| ASPO-Math-8BBackbone=Qwen3-8B2025.10 | 38.9 | 80 | |
| GRPOBackbone=Qwen3-4B2025.10 | 35.5 | 70 | |
| ASPO-Math-1.5BBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.10 | 35.1 | 70 | |
| GRPOBackbone=Qwen3-8B2025.10 | 34.1 | 66.7 | |
| Nemotron-1.5BBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.10 | 33.1 | 60 | |
| DeepScaleR-1.5BBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.10 | 29 | 63.3 | |
| GRPOBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.10 | 28.6 | 56.7 | |
| DeepSeek-R1-1.5BBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.10 | 23.5 | 63.3 | |
| Qwen3-30B-A3BBackbone=Qwen3-30B-A3B2025.10 | 19.9 | 56.7 | |
| Qwen3-8BBackbone=Qwen3-8B2025.10 | 19.1 | 56.7 | |
| Qwen3-4BBackbone=Qwen3-4B2025.10 | 18.3 | 63.3 |