Mage-VL: An Efficient Codec-Native Streaming Multimodal Foundation Model

Open paper
Benchmark Model Rank Results
infographic-visual-question-answering-on-infovqaMage-VL-4B#5ANLS: 80.33
visual-question-answering-vqa-on-ai2d-test-with-maskMage-VL-4B#6Accuracy: 83.16
visual-question-answering-on-docvqa-testMage-VL-4B#11ANLS: 0.9514
video-question-answering-on-tempcompass-overallMage-VL-4B (tc16)#15Accuracy: 62.4
video-question-answering-on-mlvuMage-VL-4B (tc32)#39Accuracy: 68.7
chart-question-answering-on-chartqaMage-VL-4B#461:1 Accuracy: 84.88
optical-character-recognition-ocr-on-ocrbenchMage-VL-4B#48Score: 818
video-question-answering-on-lvbenchMage-VL-4B (tc32)#48Accuracy: 41.8
video-question-answering-on-mvbenchMage-VL-4B (tc16)#50Avg.: 65.5
vision-language-understanding-on-mmstarMage-VL-4B#50Accuracy: 67.32

Rank counts only results with a code link.