Mathematical Reasoning on AIME 2025 (Score, Best Score, ΔH)
0.39ScoreRLOO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| RLOOBackbone=Qwen-3-32B, Context length=4K2026.03 | 0.39 | 0.4 | 0.04 | |
| GRPOBackbone=Qwen-3-32B, Context length=4K2026.03 | 0.36 | 0.39 | 0.06 | |
| REPO-RBackbone=Qwen-3-32B, Context length=4K2026.03 | 0.35 | 0.38 | 0.17 | |
| ADAPOBackbone=Qwen-3-32B, Context length=4K2026.03 | 0.34 | 0.35 | 0.29 | |
| DAPOBackbone=Qwen-3-32B, Context length=4K2026.03 | 0.31 | 0.35 | 0.75 |