Mathematical Reasoning on AIME25 (Accuracy, Performance Gain)
70Accuracyl2-guided Selection
Evaluation results
| Method | Links | ||
|---|---|---|---|
| l2-guided SelectionModel=Qwen3-14B2026.06 | 70 | 5 | |
| l2-guided SelectionModel=Qwen3-32B2026.06 | 70 | 1.2 | |
| Base (Mean@8)Model=Qwen3-32B2026.06 | 69.17 | — | |
| Base (Mean@8)Model=Qwen3-14B2026.06 | 66.67 | — | |
| l2-guided SelectionModel=Qwen3-8B2026.06 | 63.33 | 5.55 | |
| l2-guided SelectionModel=Qwen3-4B2026.06 | 60 | 5.89 | |
| Base (Mean@8)Model=Qwen3-8B2026.06 | 60 | — | |
| l2-guided SelectionModel=LLaMA-70B2026.06 | 60 | 7.47 | |
| Base (Mean@8)Model=Qwen3-4B2026.06 | 56.66 | — | |
| Base (Mean@8)Model=LLaMA-70B2026.06 | 55.83 | — | |
| l2-guided SelectionModel=Qwen3-1.7B2026.06 | 36.67 | 18.94 | |
| l2-guided SelectionModel=LLaMA-8B2026.06 | 36.67 | 10.02 | |
| Base (Mean@8)Model=LLaMA-8B2026.06 | 33.33 | — | |
| Base (Mean@8)Model=Qwen3-1.7B2026.06 | 30.83 | — |