PaperBench: Assessing How Well AI Can Reproduce AI Research
PaperBench is a benchmark designed to evaluate how well AI agents can replicate state-of-the-art AI research. Unlike traditional quiz-style evaluations, replication demands reading dense papers, recon…