A recent study by research firm Mercur (spelled Mercor in the source) compared 12 licensed CPAs with several AI models on a set of bookkeeping tasks. The AI models solved the tasks almost flawlessly, beating the accountants on speed, accuracy and cost. However, the models still could not complete a full set of accounting work without a human supervising the final steps.
What the test showed
The benchmark used simplified tasks from the APEX Accounting Benchmark, a collection of 160 tasks across 10 simulated companies. The best model, Claude Opus 5.5, met 61.8 % of the grading criteria, followed closely by Fable 5.1 at 61.0 % and GPT‑6 Astra at 57.9 %. In the same test, the average CPA scored about 37 % when working without AI assistance. The study notes that none of the models reached the 60 % threshold needed to close the books on their own.
Why accountants are still needed
Mercor points out that the benchmark focused on tasks AI does well—finding details and following exact instructions. It left out parts of the job that require human judgment, such as talking with clients, coordinating with colleagues, and applying years of contextual knowledge. Because of these gaps, accountants remain essential for final review and decision‑making, even if AI can handle the bulk of data‑driven work.
Why it matters
For small‑business owners, AI can speed up routine bookkeeping and lower costs, but a human accountant is still required to review and finalize the books. This means firms can expect productivity gains, yet they must budget for professional oversight to ensure compliance and accuracy.
