Mathematics on MATH 500 (avg@32 accuracy)
83.3Accuracy (avg@32)IBTPO
Evaluation Results
| Method | Links | |
|---|---|---|
| IBTPOBase Model=Qwen3-8B-Base2026.05 | 83.3 | |
| TreeRLBase Model=Qwen3-8B-Base2026.05 | 82.5 | |
| TreePOBase Model=Qwen3-8B-Base2026.05 | 82.2 | |
| IBROBase Model=Qwen3-8B-Base2026.05 | 82 | |
| GRPO w/ Entropy Reg†Base Model=Qwen3-8B-Base2026.05 | 81.8 | |
| Vanilla GRPOBase Model=Qwen3-8B-Base2026.05 | 81.5 | |
| GRPO w/ Clip-higherBase Model=Qwen3-8B-Base2026.05 | 81.2 | |
| IBTPOBase Model=Qwen3-1.7B-Base2026.05 | 70.1 | |
| TreeRLBase Model=Qwen3-1.7B-Base2026.05 | 67.2 | |
| Vanilla GRPOBase Model=Qwen3-1.7B-Base2026.05 | 66.8 | |
| TreePOBase Model=Qwen3-1.7B-Base2026.05 | 66.6 | |
| IBROBase Model=Qwen3-1.7B-Base2026.05 | 66.4 | |
| GRPO w/ Clip-higherBase Model=Qwen3-1.7B-Base2026.05 | 65.8 | |
| Initial ModelBase Model=Qwen3-8B-Base2026.05 | 65.7 | |
| GRPO w/ Entropy Reg†Base Model=Qwen3-1.7B-Base2026.05 | 64.8 | |
| Initial ModelBase Model=Qwen3-1.7B-Base2026.05 | 51.2 |