Math on AIME24 (Avg@16)
76.5Avg@16Base
Evaluation Results
| Method | Links | |
|---|---|---|
| BaseBackbone=Qwen3-8B (w/ thinking), Traj.=–2026.06 | 76.5 | |
| +Forward KL w/ ClipBackbone=Qwen3-8B (w/ thinking), Traj.=yo2026.06 | 76.5 | |
| TRDBackbone=Qwen3-8B (w/ thinking), Traj.=yr2026.06 | 76.5 | |
| +Reverse KL w/ Top-KBackbone=Qwen3-8B (w/ thinking), Traj.=yo2026.06 | 75.6 | |
| +Reverse KLBackbone=Qwen3-8B (w/ thinking), Traj.=yo2026.06 | 75.4 | |
| +Forward KLBackbone=Qwen3-8B (w/ thinking), Traj.=yo2026.06 | 74.8 | |
| BaseBackbone=Qwen3-4B-Instruct-2507, Traj.=–2026.06 | 63.1 | |
| TRDBackbone=Qwen3-4B-Instruct-2507, Traj.=yr2026.06 | 63.1 | |
| +Reverse KL w/ Top-KBackbone=Qwen3-4B-Instruct-2507, Traj.=yo2026.06 | 63 | |
| +Forward KL w/ ClipBackbone=Qwen3-4B-Instruct-2507, Traj.=yo2026.06 | 62.1 | |
| +Forward KLBackbone=Qwen3-4B-Instruct-2507, Traj.=yo2026.06 | 60.3 | |
| +Reverse KLBackbone=Qwen3-4B-Instruct-2507, Traj.=yo2026.06 | 58.4 | |
| AIPOModel=Qwen2.5-32B-Instruct2026.05 | 39.06 | |
| GRPOModel=Qwen2.5-32B-Instruct2026.05 | 35.71 | |
| SFTModel=Qwen2.5-32B-Instruct2026.05 | 32.61 | |
| OriginalModel=Qwen2.5-32B-Instruct2026.05 | 29.88 |