I present and analyze the case of COVID-19 modeling at the Public Health Agency of Sweden (FoHM) between February 2020 and May 2021. The analysis casts the case as a decision problem: modelers choose from a strategically prepared menu that model which they have reasons to believe will best serve their current purpose. Specifically, I argue that the model choice at FoHM concerned a trade-off between model-target similarity and model simplicity. Five reasons for choosing to engage in such a trade-off are discussed: lack of information, avoiding overfitting, avoiding fuzzy modularity, maintaining good communication, and facilitating error avoidance and detection. I conclude that the case illustrates that model simplicity is an epistemically important principle.