ATLAS Finance test reportedly puts best AI at 12%, humans at 100%

A post describing ATLAS Finance testing reports “Human: 100%. Best AI: 12%.” The author says 11 frontier models received 100 expert-level tasks, each taking humans 15–30 hours. Agents had access to data rooms, email, chat, calendars, documents, notes and Excel. According to the account, finance professionals built dozens of coworker and client personas to introduce ambiguity, controlled contradictions and real-time updates that agents had to resolve to produce client-ready work.