Mathematical Reasoning on Contest Suite (AMC23, AMC24, AIME24-26, HMMT25, BRUMO25) (test)
89.1AMC23 ScorePOISE
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| POISEModel=Qwen3-4B2026.05 | 89.1 | 59.2 | 46.9 | 43.7 | 44.3 | 28 | 38.7 | 50 | |
| DAPOModel=Qwen3-4B2026.05 | 87.6 | 60.7 | 49 | 47.5 | 45.7 | 26.7 | 38.4 | 50.8 | |
| DAPOModel=Deepseek-Distill-Qwen-1.5B2026.05 | 69.7 | 44.7 | 25.4 | 21.9 | 19.8 | 6.5 | 19.1 | 29.6 | |
| POISEModel=Deepseek-Distill-Qwen-1.5B2026.05 | 69.4 | 44.6 | 27 | 23.4 | 21.3 | 6.6 | 19.8 | 30.3 | |
| baseModel=Qwen3-4B2026.05 | 42.2 | 31.9 | 26.3 | 19.6 | 24.4 | 12.9 | 21.7 | 25.8 | |
| baseModel=Deepseek-Distill-Qwen-1.5B2026.05 | 16.9 | 7.8 | 6.7 | 6.7 | 10.4 | 2.1 | 4.2 | 7.8 |