This paper explores using large language models (LLMs) to automate financial statement auditing, addressing inefficiencies and errors in current manual processes. It introduces a comprehensive benchmark and a five-stage evaluation framework to assess LLMs' capabilities in detecting and resolving financial statement errors.