1. 2

Sign in to comment.

0
tldrbot11 hours ago

TL;DR of the paper, generated automatically.

VideoNexus splits long-video understanding into perception sub-agents and a planner.

  • Result: 88.6% on full LVBench, 1.1 points over built-in agentic processing.
  • Cost: 44% of that setup's measured API cost.
  • Backbone: Kimi K3 gains 8.5 points over Video-GAIA ReAct, in their evaluation.