| mathematical-reasoning-on-aime-2025 | ZPPO (Qwen3.5-2B) | – | Accuracy: 26.7 |
| mathematical-reasoning-on-aime-2026 | ZPPO (Qwen3.5-2B) | – | Accuracy: 40.0 |
| multi-task-language-understanding-on-mmlu-pro | ZPPO (Qwen3.5-2B) | – | Accuracy: 68.2 |
| multimodal-reasoning-on-mmmu-pro | ZPPO (Qwen3.5-2B) | – | Accuracy: 53.2 |
| multiple-choice-question-answering-mcqa-on-gpqa-diamond | ZPPO (Qwen3.5-2B) | – | Accuracy: 60.6 |
| question-answering-on-humanity-s-last-exam-hle | ZPPO (Qwen3.5-2B) | – | Accuracy: 5.1 |
| video-question-answering-on-mmvu | ZPPO (Qwen3.5-2B) | – | Accuracy: 59.0 |
| video-question-answering-on-mvbench | ZPPO (Qwen3.5-2B) | – | Avg.: 62.7 |
| visual-question-answering-on-mm-vet | ZPPO (Qwen3.5-2B) | – | GPT-4 score: 77.1 |