Kling-Keye (Yancheng Long et al.) · Video Agent Framework

VideoNexus: Video Perception Beyond the Reasoning Context

VideoNexus is a training-free framework for long-video understanding that separates detailed perception from planning. A planner decides where to inspect, what to resolve and how densely to sample, while perception sub-agents process native audio-video in independent contexts and return compact, temporally grounded evidence. Using Gemini 3.8 Flash for both planning and perception, it reports 88.6% accuracy on full LVBench, 3.6 points above same-backbone static ingestion and 1.1 points above built-in agentic processing, at 44% of the latter's measured API cost. With Kimi K3 as the backbone, it improves 8.5 points over the official Video-GAIA ReAct harness in the authors' evaluation.

Open the source ↗

Similar Video & Animation releases

Published by Quanta AI Labs, an AI agency in India. Browse every week in AI News Weekly.