ProgramBench

About

A benchmark that challenges AI agents to rebuild complete programs from scratch using only compiled binaries and documentation. Tests whether language models can reverse-engineer and implement working codebases that reproduce original program behavior.