Kwai Keye-VL 1.5 Technical Report

Benchmark Model Rank Results
video-question-answering-on-tempcompass-overallKeye-VL-1.5-8B#2Accuracy: 75.5
visual-question-answering-vqa-on-mmbench-enKeye-VL-1.5-8B#2Accuracy: 92.0
multimodal-reasoning-on-visulogicKeye-VL-1.5-8B#9Accuracy: 23.1
vision-language-understanding-on-mmstarKeye-VL-1.5-8B#10Accuracy: 80.5
visual-question-answering-vqa-on-hallusionbench-average-accuracyKeye-VL-1.5-8B#12Accuracy: 62.7
visual-question-answering-vqa-on-simplevqaKeye-VL-1.5-8B#13Accuracy: 42.9
zero-shot-video-question-answer-on-video-mme-w-o-subsKeye-VL-1.5-8B#15Accuracy (%): 73.0
video-question-answering-on-mmvuKeye-VL-1.5-8B#16Accuracy: 68.3
video-question-answering-on-video-mmmuKeye-VL-1.5-8B#20Accuracy: 66.0
optical-character-recognition-ocr-on-ocrbenchKeye-VL-1.5-8B#22Score: 866
multimodal-reasoning-on-mathvistaKeye-VL-1.5-8B#23Accuracy: 81.2
multimodal-reasoning-on-mmmuKeye-VL-1.5-8B#27Accuracy: 71.4
visual-question-answering-vqa-on-realworldqaKeye-VL-1.5-8B#31Accuracy: 73.5
multimodal-reasoning-on-mathvisionKeye-VL-1.5-8B#32Accuracy: 46.8
visual-question-answering-vqa-on-blinkKeye-VL-1.5-8B#38Accuracy: 54.9