HUMAN BASELINES FOR BENCHMARKS: AI NOW OUTPERFORMS JUNIOR ACCOUNTANTS

Summary from the TLDR Fintech newsletter

A study of 12 CPAs found that frontier AI models were faster, more accurate, and dramatically cheaper than humans on four structured month-end close tasks, with Claude Opus 5 scoring 100% while accountants averaged about 37%. Mercor cautions that this does not mean accountants are broadly replaceable, since the benchmark favors detail-heavy file search and instruction following while excluding skills such as client communication, judgment, and accumulated business context, but it suggests substantial automation potential in structured accounting work.