Study: AI beats licensed CPAs on structured accounting tasks

On October 2, 2026, The Decoder reported on a study by Mercor examining how AI models perform on accounting work. The study found that current AI models outperform licensed CPAs on structured accounting tasks in both speed and accuracy. That marks a reversal from 18 months earlier, when models of this kind still lagged far behind on such work. The report does not name the specific models tested, and it does not list task counts or accuracy figures.

The study also put models through a more demanding evaluation called the APEX Benchmark. On APEX, no model fully completed all of the tasks. According to the report, without human oversight, AI still cannot close the books on its own. The results therefore split into two parts: AI has moved ahead of licensed CPAs on structured accounting tasks, but the harder benchmark shows it is not yet able to carry out an end-to-end accounting close without a person supervising the process.

The APEX Benchmark is described as more demanding than the structured tasks used for the CPA comparison. The study's headline result, that AI beats licensed CPAs on speed and accuracy, is therefore limited to that task type. The report does not claim AI can replace accountants; it states that human oversight remains necessary for closing the books.

The story currently sits at the publication of Mercor's findings. The Decoder's article does not indicate whether the study has been peer-reviewed, nor does it say whether a full paper or a broader round of testing will follow. It also does not address how the models were selected or which specific accounting tasks were included.