The thesis deals with elementary school mathematics problems. Specifically, it will use data from the online learning environment Umíme (umimeto.org). The primary goal of the thesis is to develop and evaluate methods for the prediction of task difficulty from the task statement. The focus will be on interpretable methods that can provide actionable insight into task difficulty. The secondary goal of the thesis is to explore abilities of current large language models to solve the studied tasks and to automatically generate explanations.