Frontier models solve only 56.9% of expert European executive tasks
EuroExec contains 413 open-ended tasks written by 47 experts using more than 4,000 evaluation hours; expert answers were preferred over model outputs in 74% of direct comparisons, while automatic evaluators missed important quality differences.
Why it matters: Strong general benchmarks do not establish professional judgment on ambiguous work built from real cases, incomplete information, and consequential trade-offs.