Evaluating an AI model and evaluating an AI agent are related—but they answer fundamentally different questions. A model benchmark tests the capability of a...