Mathematical Reasoning on BeyondAIME (pass@16)
69Pass@16Reverse KL w/ Top-K
Evaluation Results
| Method | Links | |
|---|---|---|
| Reverse KL w/ Top-KBackbone=Qwen3-8B (w/ thinking), Trajectory=yo2026.06 | 69 | |
| Forward KLBackbone=Qwen3-8B (w/ thinking), Trajectory=yo2026.06 | 68 | |
| TRDBackbone=Qwen3-8B (w/ thinking), Trajectory=yr2026.06 | 68 | |
| BaseBackbone=Qwen3-8B (w/ thinking), Trajectory=–2026.06 | 66 | |
| TRDBackbone=Qwen3-4B-Instruct-2507, Trajectory=yr2026.06 | 64 | |
| Reverse KLBackbone=Qwen3-8B (w/ thinking), Trajectory=yo2026.06 | 64 | |
| Forward KL w/ ClipBackbone=Qwen3-4B-Instruct-2507, Trajectory=yo2026.06 | 62 | |
| TRDModel Backbone=Qwen3-4B-Instruct-2507, Trajectory Type (Traj.)=yr, Teacher Model=Qwen3-8B2026.06 | 62 | |
| Forward KL w/ ClipBackbone=Qwen3-8B (w/ thinking), Trajectory=yo2026.06 | 61 | |
| Forward KLModel Backbone=Qwen3-4B-Instruct-2507, Trajectory Type (Traj.)=yo, Teacher Model=Qwen3-8B2026.06 | 61 | |
| Forward KL w/ ClipModel Backbone=Qwen3-4B-Instruct-2507, Trajectory Type (Traj.)=yo, Teacher Model=Qwen3-8B2026.06 | 61 | |
| Reverse KLBackbone=Qwen3-4B-Instruct-2507, Trajectory=yo2026.06 | 60 | |
| BaseBackbone=Qwen3-4B-Instruct-2507, Trajectory=–2026.06 | 59 | |
| BaseModel Backbone=Qwen3-4B-Instruct-2507, Trajectory Type (Traj.)=-, Teacher Model=Qwen3-8B2026.06 | 59 | |
| Reverse KLModel Backbone=Qwen3-4B-Instruct-2507, Trajectory Type (Traj.)=yo, Teacher Model=Qwen3-8B2026.06 | 59 | |
| Forward KLBackbone=Qwen3-4B-Instruct-2507, Trajectory=yo2026.06 | 58 | |
| Reverse KL w/ Top-KBackbone=Qwen3-4B-Instruct-2507, Trajectory=yo2026.06 | 57 | |
| Reverse KL w/ Top-KModel Backbone=Qwen3-4B-Instruct-2507, Trajectory Type (Traj.)=yo, Teacher Model=Qwen3-8B2026.06 | 57 | |
| Forward KL w/ ClipModel Backbone=Qwen3-1.7B (w/ thinking), Trajectory Type (Traj.)=yo, Teacher Model=Qwen3-8B2026.06 | 47 | |
| BaseModel Backbone=Qwen3-1.7B (w/ thinking), Trajectory Type (Traj.)=-, Teacher Model=Qwen3-8B2026.06 | 45 | |
| TRDModel Backbone=Qwen3-1.7B (w/ thinking), Trajectory Type (Traj.)=yr, Teacher Model=Qwen3-8B2026.06 | 45 | |
| Forward KLModel Backbone=Qwen3-1.7B (w/ thinking), Trajectory Type (Traj.)=yo, Teacher Model=Qwen3-8B2026.06 | 44 | |
| Reverse KL w/ Top-KModel Backbone=Qwen3-1.7B (w/ thinking), Trajectory Type (Traj.)=yo, Teacher Model=Qwen3-8B2026.06 | 44 | |
| Reverse KLModel Backbone=Qwen3-1.7B (w/ thinking), Trajectory Type (Traj.)=yo, Teacher Model=Qwen3-8B2026.06 | 41 | |
| TRDModel=Qwen3-4B-Instruct-2507, Traj.=yr2026.06 | 32.6 | |
| Forward KL w/ ClipModel=Qwen3-4B-Instruct-2507, Traj.=yo2026.06 | 32.4 | |
| BaseModel=Qwen3-4B-Instruct-2507, Traj.=–2026.06 | 32.2 | |
| Forward KLModel=Qwen3-4B-Instruct-2507, Traj.=yo2026.06 | 32 | |
| Reverse KLModel=Qwen3-4B-Instruct-2507, Traj.=yo2026.06 | 31.6 | |
| Reverse KL w/ Top-KModel=Qwen3-4B-Instruct-2507, Traj.=yo2026.06 | 31.2 | |
| S-trace-0.9Backbone=Qwen3-8B2026.05 | 27.84 | |
| GRPO(λ)-0.9Backbone=Qwen3-8B2026.05 | 27.61 | |
| Reverse KL w/ Top-KModel=Qwen3-1.7B (w/ thinking), Traj.=yo2026.06 | 21.3 | |
| Forward KLModel=Qwen3-1.7B (w/ thinking), Traj.=yo2026.06 | 21.2 | |
| Forward KL w/ ClipModel=Qwen3-1.7B (w/ thinking), Traj.=yo2026.06 | 20.8 | |
| BaseModel=Qwen3-1.7B (w/ thinking), Traj.=–2026.06 | 20.1 | |
| TRDModel=Qwen3-1.7B (w/ thinking), Traj.=yr2026.06 | 20.1 | |
| Reverse KLModel=Qwen3-1.7B (w/ thinking), Traj.=yo2026.06 | 19.7 | |
| GRPOBackbone=Qwen3-8B2026.05 | 17.68 |