ARC-AGI leaderboard thread splits on whether the Opus 5 gap is real

Sources: ARC Prize leaderboard, HN 49045040
Summary: The ARC Prize leaderboard reached the Hacker News top of the day on 2026-07-25 with 168 points. Commenters note the size of the gap between Opus 5 and the next model on ARC-AGI 3 and ask why one vendor keeps leading these boards while their own day-to-day coding experience does not shift. Recurring objections are that any benchmark stops measuring generalization once labs can access it freely and have time to train against it, and that puzzle-style tasks may now be in training sets. Others ask why Fable 5 is absent from the board. No commenter presents evidence of contamination.
Why it matters: The leaderboard is the source for the ARC-AGI 3 claim in Anthropic's Opus 5 announcement, and the thread is a reminder that a public leaderboard's value decays once it is a target.