Mathematical Reasoning on AIME 24 (avg@16, pass@32)
35Avg@16ISPO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ISPOBackbone=Qwen2.5-7B2026.06 | 35 | 67 | |
| ProGRPOBackbone=Qwen2.5-7B2026.06 | 26.7 | 53.3 | |
| Scaf-GRPOBackbone=Qwen2.5-7B2026.06 | 22.5 | 50 | |
| Vanilla GRPOBackbone=Qwen2.5-7B2026.06 | 14.4 | 46.7 |