CVPR 2026

SAGE: Training Smart Any-Horizon Agents for Long Video Reasoning with Reinforcement Learning