← All news·2026-10-02·3 min read

Artificial intelligence surpasses accountants but won’t replace them.

Mercor benchmarked AI agents on a set of accounting problems. Humans performed poorly (around 37%) on simple tasks, whereas AI models performed almost perfectly. Claude Opus 5.5 achieved the highest score on the entire APEX Benchmark (61.8% of criteria satisfied).

aibenchmarks

What’s interesting is that Mercor benchmarked modern AI agents on a set of accounting problems solved by 12 certified professionals with an average of 5.5 years of experience. In the simple test, humans were wrong almost always. They achieved around 37% accuracy. Meanwhile, AI models solved the same tasks almost error-free. In the complete APEX Benchmark, consisting of 160 tasks divided across 10 different companies, Claude Opus 5.5 performed best, achieving 61.8% of criteria satisfied. However, even Opus did not finish almost 60% of all tasks in the benchmark completely.

This is consistent with how I use Claude Code. I use Opus 5 daily, Fable 5.1 for difficult tasks, and GPT-6 Astra is roughly equivalent to Opus in my opinion, but slightly behind Fable. In the benchmark, Astra achieved 57.9%, Opus achieved 61.8%, and Fable achieved 61.0%. So don’t panic and start laying off accountants just yet. This benchmark excluded client interaction and experience-based decision making, which are crucial parts of the job. Before using any agent on your sensitive data, make sure to test it on a sample of your own invoices and verify the errors manually instead of relying on the results from this benchmark.

Source: the-decoder.com

Free course

Stop reading about AI — start building with it

The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.

Start free →
EAEvgenii Arsentev

Author

Evgenii Arsentev

PhD · Founder & CEO, Dad4Kids