ArXiv 2025

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