UniG2U-Bench: Do Unified Models Advance Multimodal Understanding? Paper • 2603.03241 • Published Mar 3 • 87