Although several NLP benchmarks exist, few are tailored to specialized domains. CUAD, recently OSS’ed by The Atticus Project, addresses this gap in the legal domain by providing a new dataset to evaluate contract review models (which are trained to identify critical portions of a contract that humans should consider). The CUAD dataset includes more than 13,000 labels from experienced lawyers in 510 commercial contracts to identify 41 types of important legal clauses. The repository also provides checkpoints for three Transformer models fine-tuned on CUAD.