Mathematical Reasoning on AIME25 (maj@16, pass@16)
86.7Pass@16Forward KL w/ Clip
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Forward KL w/ ClipBackbone=Qwen3-8B (w/ thinking), Trajectory=yo2026.06 | 86.7 | — | |
| Reverse KLBackbone=Qwen3-8B (w/ thinking), Trajectory=yo2026.06 | 86.7 | — | |
| TRDBackbone=Qwen3-8B (w/ thinking), Trajectory=yr2026.06 | 86.7 | — | |
| Reverse KL w/ Top-KBackbone=Qwen3-8B (w/ thinking), Trajectory=yo2026.06 | 86.6 | — | |
| BaseBackbone=Qwen3-8B (w/ thinking), Trajectory=–2026.06 | 83.3 | — | |
| Forward KLBackbone=Qwen3-8B (w/ thinking), Trajectory=yo2026.06 | 83.3 | — | |
| PPCBackbone=Qwen3-4B2026.05 | 80 | 63.33 | |
| TRDBackbone=Qwen3-4B-Instruct-2507, Trajectory=yr2026.06 | 80 | — | |
| BaseBackbone=Qwen3-4B-Instruct-2507, Trajectory=–2026.06 | 76.7 | — | |
| Forward KLBackbone=Qwen3-4B-Instruct-2507, Trajectory=yo2026.06 | 76.7 | — | |
| Forward KL w/ ClipBackbone=Qwen3-4B-Instruct-2507, Trajectory=yo2026.06 | 76.7 | — | |
| Reverse KLBackbone=Qwen3-4B-Instruct-2507, Trajectory=yo2026.06 | 76.7 | — | |
| Reverse KL w/ Top-KBackbone=Qwen3-4B-Instruct-2507, Trajectory=yo2026.06 | 76.7 | — | |
| Prompt-OnlyBackbone=Qwen3-4B2026.05 | 76.67 | 60 | |
| GRPOBackbone=Qwen3-4B2026.05 | 76.67 | 60 | |
| PTA-GRPOBackbone=Qwen3-4B2026.05 | 76.67 | 60 | |
| Plan-TuningBackbone=Qwen3-4B2026.05 | 76.67 | 56.67 | |
| BaseBackbone=Qwen3-4B2026.05 | 73.33 | 60 | |
| PPCBackbone=Qwen2.5-7B2026.05 | 36.67 | 20 | |
| PPCBackbone=Qwen2.5-Math-7B2026.05 | 36.67 | 23.33 | |
| GRPOBackbone=Qwen2.5-Math-7B2026.05 | 33.33 | 20 | |
| BaseBackbone=Qwen2.5-7B2026.05 | 30 | 10 | |
| Prompt-OnlyBackbone=Qwen2.5-7B2026.05 | 30 | 16.67 | |
| GRPOBackbone=Qwen2.5-7B2026.05 | 30 | 16.67 | |
| BaseBackbone=Qwen2.5-Math-7B2026.05 | 30 | 20 | |
| Prompt-OnlyBackbone=Qwen2.5-Math-7B2026.05 | 30 | 16.67 | |
| Plan-TuningBackbone=Qwen2.5-7B2026.05 | 26.67 | 20 | |
| Plan-TuningBackbone=Qwen2.5-Math-7B2026.05 | 26.67 | 16.67 | |
| PTA-GRPOBackbone=Qwen2.5-7B2026.05 | 23.33 | 16.67 | |
| PTA-GRPOBackbone=Qwen2.5-Math-7B2026.05 | 23.33 | 16.67 | |
| PPCBackbone=Llama3.1-8B2026.05 | 20 | 6.67 | |
| Plan-TuningBackbone=Llama3.1-8B2026.05 | 13.33 | 0 | |
| GRPOBackbone=Llama3.1-8B2026.05 | 10 | 0 | |
| PTA-GRPOBackbone=Llama3.1-8B2026.05 | 10 | 6.67 | |
| BaseBackbone=Llama3.1-8B2026.05 | 3.33 | 0 | |
| Prompt-OnlyBackbone=Llama3.1-8B2026.05 | 3.33 | 0 |