DeepPHY: Benchmarking Agentic VLMs on Physical Reasoning Paper • 2508.05405 • Published Aug 7, 2025 • 64