Mercor study finds AI models beat licensed accountants at structured bookkeeping, but can't yet close the books alone
On the full APEX Accounting benchmark, Claude Opus 5.5 leads with 61.8 percent of criteria met, and no model fully solved almost 60 percent of tasks.
A study by Mercor finds that AI models are now faster, more accurate and far cheaper than licensed accountants at structured bookkeeping tasks. Twelve CPAs with an average of five and a half years of experience worked through simplified tasks from the APEX Accounting Benchmark. Eighteen months ago, the best models scored below the accountants' average of about 37 percent. Today, they solve the same tasks almost flawlessly. The full benchmark is much larger, with 160 tasks across 10 simulated companies built by more than 40 professionals averaging 11 years of experience. On it, Claude Opus 5.5 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. Mercor says no model fully solved almost 60 percent of the tasks, and that models cannot yet close the books without oversight. The company also acknowledges that the study's tasks test what AI does best, such as hunting down details and following instructions precisely. It left out key parts of the job, including talking with clients and checking in with colleagues. The results point to fast gains on narrow accounting work while showing how far full automation of the profession still is.