Mathematical Deduction on AIME25 (test)
71.11AccuracyInjectRLOpt
Evaluation Results
| Method | Links | |
|---|---|---|
| InjectRLOptbackbone=Qwen3-32B2026.02 | 71.11 | |
| InjectCorrectbackbone=Qwen3-32B2026.02 | 70 | |
| Baselinebackbone=Qwen3-32B2026.02 | 69.33 | |
| InjectRLOptbackbone=Qwen3-14B2026.02 | 67.33 | |
| InjectRLOptbackbone=Qwen3-8B2026.02 | 66 | |
| InjectCorrectbackbone=Qwen3-14B2026.02 | 64.67 | |
| InjectRLOptbackbone=Qwen3-4B2026.02 | 64 | |
| InjectCorrectbackbone=Qwen3-8B2026.02 | 61.33 | |
| Baselinebackbone=Qwen3-14B2026.02 | 59.33 | |
| Baselinebackbone=Qwen3-8B2026.02 | 58.67 | |
| InjectCorrectbackbone=Qwen3-4B2026.02 | 57.33 | |
| Baselinebackbone=Qwen3-4B2026.02 | 55.33 |