Definition
An eval is a set of realistic inputs with a way of scoring the outputs: exact checks, rules, human ratings, or another model acting as a judge. Good eval sets cover common cases, known failures and the edge cases that matter to the business.
Evals are to AI what automated tests are to software. Without them, teams cannot tell whether a new model, prompt or retrieval change made things better or worse, and quality drifts silently. Leading providers publish guidance on building them, and many teams treat the eval set as one of their most valuable assets.