Automated brain tumor segmentation methods are well established, reaching performance levels with clear clinical utility. Most algorithms require four input magnetic resonance imaging (MRI) modalities, typically T1-weighted images with and without contrast enhancement, T2-weighted images, and FLAIR images. However, some of these sequences are often missing in clinical practice, e.g., because of time constraints and/or image artifacts (such as patient motion). Therefore, substituting missing modalities to recover segmentation performance in these scenarios is highly desirable and necessary for the more widespread adoption of such algorithms in clinical routine. In this work, we report the set-up of the Brain MR Image Synthesis Benchmark (BraSyn), organized in conjunction with the Medical Image Computing and Computer-Assisted Intervention (MICCAI) 2023. The objective of the challenge is to benchmark image synthesis methods that realistically synthesize missing MRI modalities given multiple available images to facilitate automated brain tumor segmentation pipelines. The image dataset is multi-modal and diverse, created in collaboration with various hospitals and research institutions.