Evaluating generative AI requires considering more than just model performance. Language, context, instructions, system usage, and behavior are all part of an evaluation that must necessarily be multidimensional.