UniG2U-Bench: Do Unified Models Advance Multimodal Understanding? - Explained Simply | ArXiv Explained