Loading page content
Meta Internal Coding BenchCoding
Meta's internal coding benchmark: 440 tasks derived from real internal pull requests, covering bug fixes, feature development, refactoring, and code cleanup; internet access is disabled, and two attempts are made per task — each task's success rate is computed first, then averaged (pass@1).Meta Internal Coding Bench is Meta's internal coding benchmark: 440 tasks derived from real internal pull requests, covering bug fixes, feature development, refactoring, and code cleanup. Tasks run in isolated sandboxes with internet access disabled, with two attempts per task; submissions are compiled and evaluated against unit tests in dedicated grading containers.
#ModelScore
5 models total