Math Reasoning on AIME 2024, AIME 2025, and HMMT 2025 Aggregate
66.9Macro Avg. AccuracyPAINT
Evaluation Results
| Method | Links | |
|---|---|---|
| PAINTBackbone=Qwen3-8B Instruct2026.04 | 66.9 | |
| OPSDBackbone=Qwen3-8B Instruct2026.04 | 64.8 | |
| PAINTBackbone=Qwen3-4B Instruct2026.04 | 64.4 | |
| GRPOBackbone=Qwen3-8B Instruct2026.04 | 64 | |
| OPSDBackbone=Qwen3-4B Instruct2026.04 | 63.6 | |
| GRPOBackbone=Qwen3-4B Instruct2026.04 | 62.7 | |
| No updateBackbone=Qwen3-8B Instruct2026.04 | 61.8 | |
| No updateBackbone=Qwen3-4B Instruct2026.04 | 61.2 | |
| SFTBackbone=Qwen3-8B Instruct2026.04 | 59.8 | |
| SFTBackbone=Qwen3-4B Instruct2026.04 | 58.6 | |
| PAINTBackbone=Qwen3-1.7B Instruct2026.04 | 44.5 | |
| OPSDBackbone=Qwen3-1.7B Instruct2026.04 | 43.4 | |
| GRPOBackbone=Qwen3-1.7B Instruct2026.04 | 37.7 | |
| No updateBackbone=Qwen3-1.7B Instruct2026.04 | 37.1 | |
| SFTBackbone=Qwen3-1.7B Instruct2026.04 | 35.8 |