| image-generation-on-wise | LVRPO | – | Overall: 0.85Cultural: 0.86Time: 0.78Space: 0.93Biology: 0.89… |
| multimodal-reasoning-on-mathvista | LVRPO (1.5B) | – | Accuracy: 68.5 |
| multimodal-reasoning-on-mathvista | LVRPO (7B) | – | Accuracy: 76.2 |
| multimodal-reasoning-on-mmmu | LVRPO (1.5B) | – | Accuracy: 47.5 |
| multimodal-reasoning-on-mmmu | LVRPO (7B) | – | Accuracy: 58.1 |
| text-to-image-generation-on-geneval | LVRPO | – | Overall: 0.91 |
| visual-question-answering-on-mm-vet | LVRPO (1.5B) | – | GPT-4 score: 53.2 |
| visual-question-answering-on-mm-vet | LVRPO (7B) | – | GPT-4 score: 69.5 |
| visual-question-answering-vqa-on-mmbench-en | LVRPO (1.5B) | – | Accuracy: 83.2 |
| visual-question-answering-vqa-on-mmbench-en | LVRPO (7B) | – | Accuracy: 87.2 |
| visual-question-answering-vqa-on-mme | LVRPO (1.5B) | – | Score: 2298 |
| visual-question-answering-vqa-on-mme | LVRPO (7B) | – | Score: 2399 |