Opus 5 clears ARC-AGI-3 levels after figuring out the rules on level 1
GregKamradt · x · 2026-07-25
A demo run of Opus 5 on ARC-AGI-3 shows a characteristic pattern: it flails on level 1 while learning the rules, then clears levels 2 and 3 without trouble.
By level 4 it pauses to try a few hypotheses, solves the problem, and keeps going smoothly. The post is framed as one of the model's most impressive runs on the public demo.
More from Models
- Google is lagging behind open-weight models on most benchmarks — burny_tech · 2026-07-25
- Claude Opus 5 tops an Artificial Analysis coding-agent benchmark at 67 — Hesamation · 2026-07-25
- Charts Show Opus 5 Peaks in Coding Performance with Medium Thinking — dejavucoder · 2026-07-25
- Opus 5, hidden-rule inference, and J-space point to a new agent stack — imjustnewatai · 2026-07-25
- LLaDA2.2 targets the real bottleneck in multi-turn agents: decode speed — Direct_Band896 · 2026-07-25
- Opus 5 was intentionally not trained on cyber tasks, but still nears Mythos 5 at finding bugs — cedric_chee · 2026-07-25