TPE handles categorical, conditional, and mixed-type hyperparameters naturally. - TPE scales to higher dimensions (50+ hyperparameters) where GPs struggle. - TPE is faster per iteration (no matrix inversion).