Kwai Keye-VL Technical Report

Benchmark Model Rank Results
visual-question-answering-vqa-on-hallusionbench-average-accuracyKeye-VL-8B#6Accuracy: 67.0
multimodal-reasoning-on-visulogicKeye-VL-8B#8Accuracy: 25.6
video-question-answering-on-tempcompass-overallKeye-VL-8B#10Accuracy: 71.5
chart-question-answering-on-charxiv-dqKeye-VL-8B#13Accuracy: 77.7
visual-question-answering-vqa-on-simplevqaKeye-VL-8B#14Accuracy: 41.6
video-question-answering-on-mmvuKeye-VL-8B#20Accuracy: 66.1
vision-language-understanding-on-mmstarKeye-VL-8B#21Accuracy: 75.5
visual-question-answering-on-v-benchKeye-VL-8B#22Accuracy: 69.6
video-question-answering-on-video-mmmuKeye-VL-8B#24Accuracy: 57.6
multimodal-reasoning-on-mathvistaKeye-VL-8B#26Accuracy: 80.7
multimodal-reasoning-on-mmmuKeye-VL-8B#26Accuracy: 71.4
zero-shot-video-question-answer-on-video-mme-w-o-subsKeye-VL-8B#31Accuracy (%): 67.7
multimodal-reasoning-on-mathvisionKeye-VL-8B#33Accuracy: 46.0
chart-question-answering-on-chartqaKeye-VL-8B#341:1 Accuracy: 86.3
optical-character-recognition-ocr-on-ocrbenchKeye-VL-8B#37Score: 851
visual-question-answering-vqa-on-blinkKeye-VL-8B#43Accuracy: 52.0
visual-question-answering-vqa-on-realworldqaKeye-VL-8B#62Accuracy: 67.7