Mathematical Reasoning on AIME 2025 (avg@4, pass@4)
59.17Average Score (avg@4)PIPO-SFT + OPD
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| PIPO-SFT + OPDBackbone=Qwen3.5-9B, Response Budget=32K-slot2026.05 | 59.17 | 83.33 | |
| RegularBackbone=Qwen3.5-9B, Response Budget=32K-slot2026.05 | 53.33 | 66.67 | |
| PIPO-SFTBackbone=Qwen3.5-9B, Response Budget=32K-slot2026.05 | 51.67 | 76.67 | |
| PIPO-SFT + OPDBackbone=Qwen3.5-4B, Response Budget=32K-slot2026.05 | 50 | 76.67 | |
| RegularBackbone=Qwen3.5-4B, Response Budget=32K-slot2026.05 | 49.17 | 63.33 | |
| Eagle-2Backbone=Qwen3.5-9B, Response Budget=32K-slot2026.05 | 49.17 | 63.33 | |
| PIPO-SFTBackbone=Qwen3.5-4B, Response Budget=32K-slot2026.05 | 42.5 | 60 | |
| Eagle-2Backbone=Qwen3.5-4B, Response Budget=32K-slot2026.05 | 40.83 | 60 | |
| MTPBackbone=Qwen3.5-9B, Response Budget=32K-slot2026.05 | 40 | 50 | |
| MTPBackbone=Qwen3.5-4B, Response Budget=32K-slot2026.05 | 34.17 | 43.33 |