Deep Agent Released R1-V: Reinforcing Super Generalization in Vision-Language Models with Cost-Effective Reinforcement Learning to Outperform Larger Models