The Medical Segmentation Decathlon (MSD) is a biomedical image analysis challenge that tests the generalizability of image segmentation algorithms across multiple tasks and modalities. The challenge aimed to investigate whether a method capable of performing well on multiple tasks would generalize well to previously unseen tasks. MSD results confirmed this hypothesis, and the winning algorithm continued to generalize well to other clinical problems for two years. Three main conclusions were drawn: (1) state-of-the-art image segmentation algorithms generalize well when retrained on unseen tasks; (2) consistent algorithmic performance across multiple tasks is a strong surrogate of algorithmic generalizability; (3) the training of accurate AI segmentation models is now commoditized to scientists not versed in AI model training. Machine learning is revolutionizing medicine, with applications ranging from disease diagnosis to antibiotic discovery. Semantic segmentation, which transforms raw medical images into clinically relevant information, is essential for clinical applications like radiotherapy planning. It is the most widely investigated medical image processing task, with about 70% of biomedical image analysis challenges dedicated to it. However, it is challenging to decide on a baseline architecture for new clinical problems. International challenges have become the de facto standard for comparative assessment of image analysis algorithms. The MSD challenge included ten different data sets with various challenging characteristics. Two phases were presented: a development phase for model development and a mystery phase to test generalization to unseen tasks. The challenge included seven known tasks and three mystery tasks. The MSD challenge's contribution was threefold: (1) organizing the first biomedical image analysis challenge with algorithms competing in multiple tasks and modalities; (2) releasing the first open framework for benchmarking medical segmentation algorithms with a focus on generalizability; (3) showing that generalization across clinical applications is possible with one single framework. The challenge involved 180 teams, with 31 submitting valid results for the development phase and 19 for the mystery phase. The winning method, nnU-Net, was extremely robust and performed well across tasks. The challenge results demonstrated that nnU-Net achieved state-of-the-art performance on many tasks, including against task-optimized networks. The MSD challenge also showed that the training of accurate semantic segmentation networks can now be fully automated, allowing computationally-versed scientists without AI-specific knowledge to use these techniques. The challenge data sets, including ten heterogeneous tasks from various body parts and regions of interest, were released with a permissive copyright license, promoting data sharing and commercial usage. The challenge results confirmed the hypothesis that a method capable of performing well on multiple tasks will generalize well to previously unseen tasks.The Medical Segmentation Decathlon (MSD) is a biomedical image analysis challenge that tests the generalizability of image segmentation algorithms across multiple tasks and modalities. The challenge aimed to investigate whether a method capable of performing well on multiple tasks would generalize well to previously unseen tasks. MSD results confirmed this hypothesis, and the winning algorithm continued to generalize well to other clinical problems for two years. Three main conclusions were drawn: (1) state-of-the-art image segmentation algorithms generalize well when retrained on unseen tasks; (2) consistent algorithmic performance across multiple tasks is a strong surrogate of algorithmic generalizability; (3) the training of accurate AI segmentation models is now commoditized to scientists not versed in AI model training. Machine learning is revolutionizing medicine, with applications ranging from disease diagnosis to antibiotic discovery. Semantic segmentation, which transforms raw medical images into clinically relevant information, is essential for clinical applications like radiotherapy planning. It is the most widely investigated medical image processing task, with about 70% of biomedical image analysis challenges dedicated to it. However, it is challenging to decide on a baseline architecture for new clinical problems. International challenges have become the de facto standard for comparative assessment of image analysis algorithms. The MSD challenge included ten different data sets with various challenging characteristics. Two phases were presented: a development phase for model development and a mystery phase to test generalization to unseen tasks. The challenge included seven known tasks and three mystery tasks. The MSD challenge's contribution was threefold: (1) organizing the first biomedical image analysis challenge with algorithms competing in multiple tasks and modalities; (2) releasing the first open framework for benchmarking medical segmentation algorithms with a focus on generalizability; (3) showing that generalization across clinical applications is possible with one single framework. The challenge involved 180 teams, with 31 submitting valid results for the development phase and 19 for the mystery phase. The winning method, nnU-Net, was extremely robust and performed well across tasks. The challenge results demonstrated that nnU-Net achieved state-of-the-art performance on many tasks, including against task-optimized networks. The MSD challenge also showed that the training of accurate semantic segmentation networks can now be fully automated, allowing computationally-versed scientists without AI-specific knowledge to use these techniques. The challenge data sets, including ten heterogeneous tasks from various body parts and regions of interest, were released with a permissive copyright license, promoting data sharing and commercial usage. The challenge results confirmed the hypothesis that a method capable of performing well on multiple tasks will generalize well to previously unseen tasks.