Google’s engineering benchmark finds top AI agents scoring just 50.6%

Google's PLE benchmark reveals top AI agents score just 50.6% on real engineering tasks, meeting hard design constraints only 41% of the time.

Published: 2026-10-28