APEX-Accounting Benchmark Evaluates Frontier Model Accounting Capabilities
July 30, 2026
APEX-Accounting assesses model performance on 160 expert-authored tasks including reconciliation, expense accrual, and report generation across 10 simulated accounting environments. Claude-Fable-5 (Max) leads with a 56.4% Mean Criteria@3, while the highest Pass@8 score across all tested models is only 21.5%.
HOW THIS AFFECTS YOU
●
builderYou can use this to evaluate how reliably your agents can handle complex financial workflows.
●
researcherThis provides a high-fidelity benchmark for testing domain-specific reasoning in specialized environments.