All terms
Evaluation
GDPval
An OpenAI benchmark measuring whether AI can do real, economically valuable professional work across many occupations.
Definition
GDPval is a benchmark from OpenAI that tests whether AI can produce the kind of work people actually get paid for. It contains over a thousand tasks designed by experienced professionals across dozens of occupations — think legal briefs, engineering documents, nursing care plans, and customer-support interactions — and judges whether a model's output meets a real-world standard. The name nods to GDP, framing it as a measure of economically useful output rather than abstract test scores, which is why it usually appears under 'knowledge work' on model comparison pages.