AI beats licensed accountants on speed and accuracy, but still can't close the books without supervision

A recent study conducted by Mercor has revealed that advanced AI models can surpass licensed accountants in both speed and accuracy when performing structured bookkeeping tasks. The research utilized the APEX Accounting Benchmark, which consists of 160 tasks across 10 simulated companies, to evaluate the performance of various AI models against licensed Certified Public Accountants (CPAs) with an average experience of 5.5 years.

The findings indicate that the AI models, specifically Claude Opus 5.5, Fable 5.1, and GPT-6 Astra, achieved accuracy rates of 61.8%, 61.0%, and 57.9%, respectively. In contrast, the licensed accountants scored an average of only 37%. This marks a significant improvement from eighteen months prior, when the best AI models scored below the accountants' average.

Despite these advancements, the study highlights a critical limitation: nearly 60% of the tasks in the benchmark were not fully solved by any of the AI models. This underscores the ongoing need for human oversight, as the AI systems struggled with essential aspects of the accounting profession, such as client communication, collaboration with colleagues, and leveraging contextual knowledge accumulated over years of experience.

Mercor's analysis suggests that while AI models are faster, more accurate, and cost-effective for structured tasks, they cannot fully replace accountants. The company anticipates that the integration of AI into the accounting industry will lead to significant productivity gains, but emphasizes that human accountants will continue to play an indispensable role in the profession.