650 AI Accountants Just Surpassed Human Accountants [What does this mean for accounting firms?]

Jason Stats of Accounting Firm Coach uses Mercor’s Apex Accounting Benchmark to argue frontier AI models are now outperforming licensed CPAs on month-end close tasks, with major implications for workflow design and labor cost. He also reviews Anthropic Claude Sonnet 5.5, ChatGPT’s new features, and why Xero has fallen behind QuickBooks Online and newer ledgers like Kick—relevant to CIOs watching AI-driven process automation and vendor shifts.

Key points

• Mercor’s benchmark found top models rose from 37% to perfect scores on simulated accounting tasks in 18 months.
• Human CPAs averaged 5.5 years’ experience and still scored below frontier models; AI was also far cheaper per task.
• The real opportunity is not replacing humans, but automating repeatable production so staff focus on relationships and advisory work.
• Claude Sonnet 5.5 is presented as fast enough and nearly as capable as Opus 5.5 for many knowledge-work tasks.
• Xero no longer ranks as the best SMB ledger option in the U.S.; QuickBooks Online and AI-native ledgers are preferred.
• Receipt management and document handling are becoming features inside broader workflows, not standalone software categories.