Artificial intelligence models have surpassed licensed accountants in speed and accuracy for structured bookkeeping tasks, according to a new study by Mercor, though the technology remains unable to independently close the books without human oversight.
What Happened
Mercor conducted a study comparing 12 licensed CPAs, who had an average of five and a half years of experience, against current AI models using simplified tasks from the APEX Accounting Benchmark. Eighteen months prior, the best AI models scored below the accountants' average of approximately 37 percent. Today, those same models solve the simplified tasks almost flawlessly.
The full APEX Accounting benchmark is more extensive, comprising 160 tasks across 10 simulated companies, developed by over 40 professionals with an average of 11 years of experience. According to the study, Claude Opus 5.5 currently leads with 61.8 percent of grading criteria met, followed by Fable 5.1 at 61.0 percent and GPT-6 Astra at 57.9 percent. However, Mercor notes that no model fully solved nearly 60 percent of the tasks in the full benchmark, indicating that AI still cannot close the books without supervision.
Why It Matters
Mercor acknowledges that the study's tasks test exactly what AI does best: hunting down details and following instructions precisely. The research excluded key aspects of the accounting profession, such as communicating with clients, collaborating with colleagues, and utilizing context accumulated over years of experience. Consequently, Mercor states that accountants cannot be replaced by these models, although the firm expects major productivity gains across the industry as AI handles structured data tasks more efficiently and cheaply than human workers.
The Bottom Line
While AI models like Claude Opus 5.5 demonstrate superior performance in speed and accuracy for structured bookkeeping compared to licensed CPAs, they lack the capability to independently manage complex accounting workflows without human supervision.