Zero AI outputs from GPT-5.4 and Claude Opus 4.6 were rated ready for client delivery in a new benchmark. Five hundred investment bankers found the results too imprecise or factually incorrect for professional use. Despite these failures, over half of the participants will still use the outputs as initial drafts. Accuracy remains a critical barrier.