Medical image segmentation is essential for accurate diagnosis, treatment planning, and disease monitoring, but traditional methods often lack generalizability across diverse modalities and diseases. Recent advancements have introduced universal models to address this challenge, achieving robust and versatile segmentation. Key models include MedSAM, trained on 1.57 million image-mask pairs across 10 modalities and 30+ cancer types, excelling in internal and external validations; MONAI Vista, which extends segmentation to 3D medical data using class-label prompts; Hermes, leveraging context-prior learning from diverse imaging datasets; UniSeg, integrating task-specific prompts for superior performance across multiple tasks; and SAT, utilizing text prompts for versatile segmentation in various scenarios. These models outperform traditional methods in accuracy and robustness, providing efficient solutions across diverse imaging tasks, accelerating diagnostics, and enabling personalized treatments. By leveraging large datasets and innovative prompts, universal models mark a significant advancement in medical image understanding and segmentation.