| image-editing-on-kris-bench | Native UMM Synergy: End-to-end UMM | – | KRIS-Bench Overall: 68.33 |
| image-editing-on-risebench | Native UMM Synergy: End-to-end UMM | – | RISEBench Overall Accuracy: 18.88 |
| mathematical-reasoning-on-mathverse | Native UMM Synergy: Und+Gen geometry | – | Accuracy: 63.15 |
| multimodal-reasoning-on-mathvista | Native UMM Synergy: Und+Gen geometry | – | Accuracy: 74.3 |
| multimodal-reasoning-on-mmmu | Native UMM Synergy: Modality-decoupled MoT U | – | Accuracy: 42.89 |
| optical-character-recognition-ocr-on-ocrbench | Native UMM Synergy: Dense U+G | – | Score: 689 |
| text-to-image-generation-on-dpg-bench | Native UMM Synergy: Modality-decoupled MoT U+G | – | DPG-Bench Score: 83.09 |
| vision-language-understanding-on-mmstar | Native UMM Synergy: Task-decoupled MoT U+G | – | Accuracy: 56.51 |
| visual-question-answering-vqa-on-blink | Native UMM Synergy: Dense U+G | – | Accuracy: 64.09 |