Mercor Says AI Models Beat 12 Licensed CPAs on Month-End Close Tasks
Summary
Mercor researchers report that frontier AI models beat 12 licensed CPAs on four month-end close scenarios drawn from the APEX-Accounting benchmark, scoring higher and costing more than ten times less per task. The firm says the tasks tested detail-oriented instruction following, not client communication, so the results do not show accountants are replaceable.
Key Points
- Mercor reports its 12 study accountants average about five and a half years of experience.
- Mercor researcher Aden Barton says eighteen months ago the best models fell short of the average accountant's roughly 37% score.
- Mercor says models alone score perfectly on the tasks, leaving no room to measure AI augmentation uplift.