Anthropic's Claude AI Outperforms Licensed Accountants on Professional Tasks in Recent Benchmark Study
Mercor's evaluation of Claude Opus 5 against 12 licensed CPAs with an average of 5.4 years of experience showed the AI model achieving 100% accuracy on realistic accounting tasks in under 10 minutes, while human accountants averaged 37% accuracy and required 30 minutes to three hours. The dramatic performance gap reflects rapid advances in AI capability over less than two years, during which frontier models have shifted from struggling to match human accountants on similar tasks to substantially exceeding them in both speed and accuracy. The results illustrate how quickly AI is reshaping professional work in domains previously considered resistant to automation.
Mercor's benchmark evaluated accountants on month-end close procedures—tasks requiring document review, data extraction, calculation, and formatted reporting. The 12 participants were experienced professionals, with most holding senior roles and half having worked at major accounting firms. The AI model completed identical assignments with perfect accuracy while incurring approximately $0.21 per successful task criterion, whereas human accountants cost around $10.35 per criterion, representing a roughly 49-fold cost differential.
The study's significance lies in the pace of advancement rather than the outcome alone. Eighteen months prior, leading AI models performed near zero on these same exercises. Within that brief window, capabilities progressed from failing to match human performance to substantially exceeding it. This trajectory demonstrates how quickly AI systems can cross professional competency thresholds that were previously considered human-resistant domains.
These findings may accelerate workforce transitions in accounting and adjacent financial services roles. Organizations could face pressure to evaluate AI integration in routine accounting operations, potentially affecting demand for entry and mid-level positions focused on standardized tasks. However, the study's authors note that results reflect narrow task categories where AI excels—file searching and instruction-following—and do not indicate wholesale accountant obsolescence. Broader implications depend on how organizations adopt these tools and whether human roles evolve toward higher-judgment activities.