| arithmetic-reasoning-on-gsm8k | NeuPAT (Qwen3-4B-Instruct-2507) (extra data) | – | Accuracy: 85.97 |
| code-generation-on-mbpp | NeuPAT (Qwen2.5-14B-Instruct) (extra data) | – | Accuracy: 66.8 |
| mathematical-reasoning-on-math-500 | NeuPAT (Qwen3-4B-Instruct-2507) (extra data) | – | Accuracy: 56.4 |
| object-hallucination-evaluation-on-pope | NeuPAT (Qwen3-4B-Instruct-2507) (extra data) | – | F1: 88.72 |
| vision-language-understanding-on-mmstar | NeuPAT (Qwen2.5-14B-Instruct) (extra data) | – | Accuracy: 50.14 |
| visual-question-answering-vqa-on-mmbench-en | NeuPAT (Qwen2.5-14B-Instruct) (extra data) | – | Accuracy: 74.31 |
| visual-question-answering-vqa-on-realworldqa | NeuPAT (Qwen3-4B-Instruct-2507) (extra data) | – | Accuracy: 56.99 |