CyberV: Cybernetics for Test-time Scaling in Video Understanding

Open paper

Benchmark	Model	Rank	Results
video-question-answering-on-video-mmmu	Qwen2.5-VL-72B + CyberV	#16	Accuracy: 74.3