Mercor recently created a new accounting benchmark called Apex-Accounting, which tasks models to perform challenging month-end close tasks using accounting software and direct access to company info and source documents. https://www.mercor.com/apex/apex-accounting-leaderboard/

The leading model, Fable 5, achieves a pass@1 score of 9.5%, and satisfies 56.4% of all criteria for the median problem.

What will the highest score be before 2027? Resolves to Mercor reporting.