AI models now outperform licensed CPAs on speed and accuracy
1 source · The Decoder- Boom: Mercor study finds AI now surpasses licensed CPAs on speed and accuracy in structured accounting tasks
- Doom: No AI model fully completes all tasks on the more demanding APEX Benchmark
- Doom: AI still cannot close financial books without human supervision, study concludes
- Neutral: Eighteen months ago, AI models still lagged behind CPAs on the same structured tasks
The story in full
A Mercor study published around October 2, 2026 found that current AI models outperform licensed CPAs on structured accounting tasks in both speed and accuracy. Eighteen months prior, the same models still lagged behind human accountants on those tasks.
On the more demanding APEX Benchmark, no model fully completes all required tasks. The study concludes that without human oversight, AI cannot independently close financial books, placing a meaningful limit on full automation of the accounting workflow.
Analysis
341 wordsMercor published a study on October 2, 2026, comparing current AI models against licensed CPAs on structured accounting tasks. The findings showed that AI now surpasses human accountants on both speed and accuracy, a significant reversal from eighteen months earlier, when the same models still lagged behind on those identical tasks. The study also tested models against the APEX Benchmark, a more demanding evaluation, and found that no model successfully completes all of its required tasks. The study's headline conclusion is that AI cannot close a company's financial books without human supervision.
The eighteen-month timeline is the detail that gives this story its weight. A capability gap that existed well into 2025 has closed entirely on structured tasks, which represents an unusually fast shift in a professional domain where licensing, liability and regulatory compliance create high barriers. What remains in dispute is how much the distinction between structured tasks and full-cycle book closing actually matters in practice. Structured accounting work covers a large share of routine accounting labor, but the book-closing process is the formal, regulated endpoint of that work, and the APEX Benchmark results suggest a meaningful ceiling remains.
No reactions from the Pro-AI, Anti-AI or Middle Ground camps have been published yet. Typically, Pro-AI voices would treat the speed and accuracy result as evidence that AI is ready to reduce headcount in accounting and that the remaining gap is an engineering problem close to being solved. Anti-AI voices would likely focus on the APEX Benchmark failure and the book-closing limitation as proof that high-stakes financial work still requires human judgment and professional accountability. The Middle Ground camp would probably argue that the right response is augmentation rather than replacement, using AI to handle structured task volume while keeping licensed professionals responsible for the regulated close process.
The argument is likely to sharpen once firms begin disclosing how they are deploying these tools in live accounting workflows, and if Mercor or an independent body updates the APEX Benchmark results to show whether any model has since closed the remaining gap.
Where do you stand?
Add your take
0 reader votesSign in with Google to pick a side and post. Your vote moves the story's Doom / Boom score.