Mathematical Reasoning In-Distribution various (test)
66.4AIME 24 ScoreTreePO
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| TreePOModel=Qwen3-14B2026.06 | 66.4 | — | — | 94.8 | 45.4 | 67.8 | 74 | 95 | |
| PCLModel=Qwen3-14B2026.06 | 66.2 | — | — | 95 | 45.1 | 67.6 | 73.9 | 95.1 | |
| TRACEModel=Qwen3-14B2026.06 | 66.1 | — | — | 94.9 | 47.3 | 71.5 | 74.9 | 94.7 | |
| GRPOModel=Qwen3-14B2026.06 | 65.6 | — | — | 94.2 | 45.2 | 68.4 | 73.5 | 94.3 | |
| TreePOModel=Qwen3-8B2026.06 | 64.3 | — | — | 92.6 | 40.9 | 63.1 | 70.4 | 91.3 | |
| PCLModel=Qwen3-8B2026.06 | 64 | — | — | 92.8 | 40.7 | 63 | 70.4 | 91.5 | |
| TRACEModel=Qwen3-8B2026.06 | 63.9 | — | — | 93.4 | 41.2 | 65.8 | 71.1 | 91.2 | |
| GRPOModel=Qwen3-8B2026.06 | 63.6 | — | — | 92.3 | 40.2 | 62.8 | 70 | 91 | |
| ReActModel=Qwen3-14B2026.06 | 61.5 | — | — | 91.8 | 40 | 62.3 | 69 | 89.4 | |
| ReActModel=Qwen3-8B2026.06 | 52.3 | — | — | 89.8 | 37.8 | 58.6 | 65.1 | 87 | |
| Oat-ZeroApproach=RLVR2025.12 | 33.4 | 11.9 | 61.2 | 78 | 34.6 | 43.4 | 43.7 | — | |
| DaGRPOOff-policy=true2025.12 | 29.5 | 20.9 | 63 | 86 | 37.5 | 55.8 | 48.8 | — | |
| SimpleRL-ZeroApproach=RLVR2025.12 | 27 | 6.8 | 54.9 | 76 | 25 | 34.7 | 37.4 | — | |
| SFT -> GRPOPipeline=Hybrid2025.12 | 25.7 | 21.6 | 62.2 | 84.6 | 38.2 | 46.8 | 46.5 | — | |
| DaGRPO (w/o Off-policy)Off-policy=false2025.12 | 23.2 | 16.1 | 63.1 | 85 | 36.7 | 51.2 | 45.9 | — | |
| LUFFYApproach=Hybrid2025.12 | 23 | 20.8 | 63.4 | 85.4 | 35.6 | 54.3 | 47.1 | — | |
| GRPOMode=On-policy RL2025.12 | 22.7 | 14.8 | 58.5 | 81.8 | 40 | 46.7 | 44.1 | — | |
| SFTMode=Fine-tuned2025.12 | 22.2 | 22.3 | 52.8 | 82.6 | 40.8 | 43.7 | 44.1 | — | |
| PRIME-ZeroApproach=RLVR2025.12 | 17 | 12.8 | 54 | 81.4 | 39 | 40.3 | 40.7 | — | |
| OpenReasoner-ZeroApproach=RLVR2025.12 | 16.5 | 15 | 52.1 | 82.4 | 33.1 | 47.1 | 41 | — | |
| Qwen2.5-Math-InstructVersion=Instruct2025.12 | 12.5 | 10.2 | 48.5 | 80.4 | 32.7 | 41 | 37.6 | — | |
| Qwen2.5-MathVersion=Base2025.12 | 10.4 | 3.3 | 29.8 | 47.6 | 9.1 | 16.1 | 19.4 | — |