One dissent: on held-out novel puzzle games the leap evaporates, since evaluations stay held-out only until someone optimizes the genre.
Development
A model quadruples the best score on the hardest reasoning benchmark
Claim
Claude Opus 5 landed at Fable-class intelligence for half the price, sweeping Frontier-Bench, GDPval and Humanity's Last Exam, and ARC Prize crowned it state of the art on ARC-AGI-3 at 30.2%, quadruple the old best, after it turned puzzle layouts into reflection equations for the first time.
Reported metric: 30.2%, 4x prior
Reporting date: (not necessarily the event date)
Sources
Attribution and reporting linked by this record; links do not imply independent verification.
- Welcome to July 26, 2026 (theinnermostloop.substack.com)
Cite or correct this record
Suggest a correction — opens a GitHub draft for you to review and submit.