Artificial Analysis's second GDPval-AA evaluation of economically valuable knowledge-work deliverables, reported as pairwise-comparison Elo.
Uses realistic professional deliverables rather than short-form questions. Publishes model comparisons together with token usage, turns, and cost context.
Elo depends on the comparison pool and evaluation setup. Scores must preserve model variant, reasoning effort, tools, and observation date and are not interchangeable with unversioned GDPval-AA imports.
Higher is better. Elo from pairwise evaluation of professional deliverables.
Elo depends on the comparison pool and evaluation setup.
No scores available for this benchmark yet.