The chance that at least one of k attempts is correct. It is the standard way of scoring code generation, because a model that gets it right one time in ten is genuinely useful when you can run the tests and keep the one that works, and useless if you can only take one shot.
Technical view
Estimated unbiasedly from n samples with c correct rather than by literally generating k. Pass@1 measures single-shot reliability; higher k measures whether a correct solution is within reach given a verifier. The gap between them quantifies how much value a selection mechanism adds, which is directly relevant to agentic systems that generate, test and retry. It requires a reliable verifier, so weak test suites inflate the measure.
Related
Part of AI Terminology Index, section: Measuring Performance.