Arxiv paper - MEGA-Bench: Scaling Multimodal Evaluation to over 500 Real-World Tasks | AI Breakdown | Podwise