| Benchmark | Model | Rank | Results |
|---|---|---|---|
| code-generation-on-scicode | Qwen3.5-35B-A3B + GRPO + Ours | – | Accuracy: 29.9 |
| mathematical-reasoning-on-aime-2026 | Qwen3.5-35B-A3B + SAO + Ours | – | Accuracy: 92.2 |
| mathematical-reasoning-on-hmmt-feb-2026 | Qwen3.5-35B-A3B + SAO + Ours | – | Accuracy: 83.3 |
| multiple-choice-question-answering-mcqa-on-gpqa-diamond | Qwen3.5-35B-A3B + GRPO + Ours | – | Accuracy: 85.0 |