Mathematical Reasoning on AIME 24 (Accuracy, Performance Gain %)
73.33Accuracyl2-guided Selection
Evaluation results
| Method | Links | ||
|---|---|---|---|
| l2-guided SelectionModel=Qwen3-32B2026.06 | 73.33 | 3.53 | |
| Base (Mean@8)Model=Qwen3-32B2026.06 | 70.83 | — | |
| l2-guided SelectionModel=Qwen3-4B2026.06 | 66.67 | 11.12 | |
| l2-guided SelectionModel=Qwen3-8B2026.06 | 66.67 | 4.71 | |
| l2-guided SelectionModel=Qwen3-14B2026.06 | 66.67 | 1.28 | |
| l2-guided SelectionModel=LLaMA-70B2026.06 | 66.67 | 5.27 | |
| Base (Mean@8)Model=Qwen3-14B2026.06 | 65.83 | — | |
| Base (Mean@8)Model=Qwen3-8B2026.06 | 63.67 | — | |
| Base (Mean@8)Model=LLaMA-70B2026.06 | 63.33 | — | |
| Base (Mean@8)Model=Qwen3-4B2026.06 | 60 | — | |
| l2-guided SelectionModel=LLaMA-8B2026.06 | 43.33 | 8.33 | |
| Base (Mean@8)Model=Qwen3-1.7B2026.06 | 40 | — | |
| l2-guided SelectionModel=Qwen3-1.7B2026.06 | 40 | 0 | |
| Base (Mean@8)Model=LLaMA-8B2026.06 | 40 | — |