Mercor puts junior accountants against frontier AI

Mercor’s October 1 human-baseline study says Claude Opus 5 scored 100% across 20 attempts on four month-end accounting tasks, finishing each in under 10 minutes. The comparison used 12 licensed CPAs averaging about 5.5 years of experience, but Mercor cautions that these well-defined tasks test only a slice of accounting work; that caveat echoes the site’s audit of thin cost evidence for AI employees. Finance teams can use the scenarios to benchmark repetitive close work, but the result does not support a staffing forecast.