Youtu-VL: Unleashing Visual Potential via Unified Vision-Language Supervision

Benchmark Model Rank Results
multimodal-reasoning-on-visulogicYoutu-VL (4B)#7Accuracy: 25.7
chart-question-answering-on-charxiv-dqYoutu-VL (4B)#12Accuracy: 79.0
mathematical-reasoning-on-mathverseYoutu-VL (4B)#14Accuracy: 56.5
object-counting-on-countbenchYoutu-VL (4B)#19Accuracy: 88.6
chart-question-answering-on-charxiv-rqYoutu-VL (4B)#27Accuracy: 44.0
visual-question-answering-vqa-on-realworldqaYoutu-VL (4B)#27Accuracy: 74.6
multi-task-language-understanding-on-mmlu-reduxYoutu-VL (4B)#30Accuracy: 76.8
multiple-choice-question-answering-mcqa-on-gpqa-diamondYoutu-VL (4B)#127Accuracy: 39.8
object-detection-on-coco-val2017Youtu-VL (4B)#269box AP: 47.1