Researchers have unveiled MARS, a large-scale foundation model designed to revolutionize multi-sequence magnetic resonance imaging (MRI) analysis by addressing heterogeneity from different anatomical structures and diverse MRI sequences. MARS employs a unique pretraining strategy that disentangles anatomy-invariant features from sequence-specific variations, allowing the model to learn robust, transferable representations. The team amassed 336,476 volumetric scans from 34 datasets encompassing 10 different anatomical regions and multiple MRI sequences, drawn from eight public and 26 private collections. To evaluate the model, researchers established a benchmark suite encompassing 44 downstream tasks spanning disease diagnosis, anatomical segmentation, spatial registration, disease progression prediction, and automated report generation. MARS secured the top ranking in 41 out of 44 challenges, often with statistically significant performance improvements over existing methods, and demonstrated strong generalization capability on external datasets differing substantially from the pretraining data.