Claude Opus 5.5 Beats CPAs on Mercor Accounting Test
A new study by Mercor reveals that top AI models now outperform human accountants in structured bookkeeping speed and accuracy, signaling a massive shift toward automated financial workflows.
Artificial intelligence has surpassed human professionals in structured bookkeeping tasks, according to a recent study conducted by Mercor. The evaluation pitted leading AI models against 12 licensed certified public accountants (CPAs) who possessed an average of five and a half years of experience. The human participants tackled simplified tasks from the APEX Accounting Benchmark. While the top AI models scored below the CPAs' average of approximately 37 percent just 18 months ago, today's frontier models complete these identical tasks with near-flawless execution.
The broader APEX Accounting benchmark is a comprehensive evaluation consisting of 160 tasks distributed across 10 simulated companies, designed by more than 40 accounting professionals with an average of 11 years of experience. In this rigorous test, Anthropic's Claude Opus 5.5 currently leads the leaderboard by meeting 61.8 percent of the grading criteria. Fable 5.1 follows closely in second place at 61.0 percent, while OpenAI's GPT-6 Astra holds the third spot with a score of 57.9 percent. Despite these high marks, Mercor noted that no model was able to fully solve nearly 60 percent of the benchmark's tasks, indicating that AI cannot yet close financial books without human oversight.
For accounting practitioners, these findings suggest a dramatic shift in daily operations rather than immediate job replacement. The Mercor study specifically highlighted structured bookkeeping, which plays to the strengths of large language models, such as retrieving specific details and adhering strictly to instructions. However, the benchmark omitted critical interpersonal and contextual aspects of the profession, including client communication, team collaboration, and the application of historical business context. Consequently, while CPAs will remain essential for high-level decision-making and relationship management, they can expect AI to handle repetitive data tasks and drive significant productivity gains across the industry.
This is our own summary of reporting by The Decoder
