Mathematical Reasoning on AIME 26 (Avg@16, Pass@16)
36.25Avg@16CAST
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CASTModel Scale=Qwen3-4B2026.05 | 36.25 | 70 | |
| CASTModel Scale=Qwen3-8B2026.05 | 32.29 | 66.7 | |
| GRPOModel Scale=Qwen3-4B2026.05 | 20 | 53.3 | |
| RLRTModel Scale=Qwen3-4B2026.05 | 20 | 63.3 | |
| BaseModel Scale=Qwen3-4B2026.05 | 17.71 | 50 | |
| RLSDModel Scale=Qwen3-4B2026.05 | 17.71 | 40 | |
| RLRTModel Scale=Qwen3-8B2026.05 | 17.71 | 53.3 | |
| CASTModel Scale=Qwen3-1.7B2026.05 | 17.29 | 36.7 | |
| GRPO+OPSDModel Scale=Qwen3-4B2026.05 | 17.29 | 50 | |
| RLSDModel Scale=Qwen3-8B2026.05 | 17.29 | 50 | |
| GRPO+OPSDModel Scale=Qwen3-8B2026.05 | 16.67 | 46.7 | |
| GRPOModel Scale=Qwen3-8B2026.05 | 15.21 | 40 | |
| BaseModel Scale=Qwen3-8B2026.05 | 14.58 | 43.3 | |
| OPSDModel Scale=Qwen3-8B2026.05 | 13.75 | 40 | |
| RLSDModel Scale=Qwen3-1.7B2026.05 | 11.04 | 33.3 | |
| GRPOModel Scale=Qwen3-1.7B2026.05 | 10 | 30 | |
| BaseModel Scale=Qwen3-1.7B2026.05 | 9.38 | 23.3 | |
| RLRTModel Scale=Qwen3-1.7B2026.05 | 9.38 | 26.7 | |
| OPSDModel Scale=Qwen3-4B2026.05 | 9.17 | 13.3 | |
| GRPO+OPSDModel Scale=Qwen3-1.7B2026.05 | 7.5 | 23.3 | |
| OPSDModel Scale=Qwen3-1.7B2026.05 | 2.92 | 13.3 |