SemiAnalysis Launches GDPx, an Agent Benchmark Where the Best Model Scores Just 12%
A new 11-domain benchmark aimed at real economic deliverables puts the top model at 12% — a brutal reality check against the saturated leaderboards agents keep topping.
SemiAnalysis has introduced a benchmark designed to measure what agents actually deliver economically, and the early results are humbling. As @wesjh_ reported, the GDPx agentic benchmark spans 11 domains, and the top-performing model scored just 12%. The number is the story: on tasks meant to approximate real-world economic output, frontier agents are failing roughly nine times out of ten.
Unlock the full briefing
Get every story in today's briefing, the full archive, and the daily AI intelligence brief.
All stories today
Full archive
Daily brief
Cancel anytime. Payments powered by Stripe.