{"id":"cybench","name":"Cybench","maker":"Stanford University researchers","url":"https://cybench.github.io","repo":"andyzorigin/cybench","summary":"A benchmark of professional-level capture-the-flag cybersecurity tasks for language model agents.","details":"Contains tasks from several recent competitions, each with a task description, a container environment and subtasks that give partial credit. It is used to measure the offensive security ability of agents. The code is open source under Apache 2.0.","open_source":true,"hosted":false,"also":["safety"],"verified":"2026-11-15","category":"benchmarks","category_name":"Agent benchmarks","page":"https://modelmoot.com/tools/cybench","reviewed":"2026-11-15","checked":{"at":"2026-11-15T14:03:44.997Z","address_answers":true,"repository":{"at":"2026-11-15T19:19:09.346Z","archived":false,"licence":"Apache-2.0","language":"HTML","last_change":"2026-10-17T07:01:27.000Z","latest_release":null}},"note":"Descriptions are the site's own. Facts under \"checked\" were read automatically from the tool's address and public repository."}