Claude Opus 5 reaches 30% on ARC-AGI-3, setting a new state of the art
typewriters · x · 2026-07-25
Fchollet says Claude Opus 5 sets a new state of the art on ARC-AGI-3, reaching 30%.
ARC-AGI-3 tests solving problems with no prior exposure, a regime where scaling has historically helped the least. The takeaway is that Opus 5 shows a notable jump on novelty-heavy reasoning, not just familiar benchmark reuse.
Related event: Anthropic Releases Claude Opus 5 with Leading Benchmarks(64 posts)→
More from Models
- Opus 5 System Card Reveals Algebra Conversion Tactic for ARC-AGI Puzzles — herbiebradley · 2026-07-25
- GPT-5.6 beat Claude Opus 5 on landing-page polish in a frontend test — Arindam_1729 · 2026-07-25
- Google is called out for weak Gemma models and its stance on open-source LLMs — BoogerheadCult · 2026-07-25
- Musk says Grok 4.6 is due in 2 weeks and Grok 4.7 in 4 weeks — elonmusk · 2026-07-25
- Together says Kimi K3 matches near-flagship coding at about 35% of Claude Fable 5’s price — togethercompute · 2026-07-25
- Musk says Grok 4.5 and Claude Opus 5 are the only models on the Pareto frontier — elonmusk · 2026-07-25