Schema Achieves High Score on ARC-AGI-3
xiuyu_l · x · 2026-07-17
This post discusses the performance of a method/framework called schema on the ARC-AGI-3 benchmark.
Key information from the original post:
- By injecting analysis-by-synthesis into the model, the authors claim to have achieved 99% RHAE on ARC-AGI-3.
- Specific results show that using Opus 4.8 + Fable 5 reaches 99%, while GPT-5.6 Sol achieves 95.35% (public set).
- The author describes this approach as enabling LLMs to "think like physicists."
The reposter summarizes it further:
- schema is essentially akin to VIGA²: performing inverse graphics first, then moving up a level for inverse dynamics.
- The core takeaway is that "analysis by synthesis" proves to be highly effective.
This is a research and benchmark-focused post, highlighting the methodology and scoring performance on ARC-AGI-3.
Related event: Schema Harness Sparks ARC-AGI-3 Debate(14 posts)→
More from Research
- Interpretability researcher mathildepapillo joins Goodfire post-PhD, calls interp the key problem — mathildepapillo · 2026-10-27
- Reasoning models can barely control their chain of thought, paper finds — tomekkorbak · 2026-10-27
- New paper: LLM data schedules and learning rate schedules should be co-designed — jasondeanlee · 2026-10-27
- What makes research get adopted: important problems, proper evals, and usability — srchvrs · 2026-10-27
- Bitter lesson part 2: few academic inventions matter, and conference publications aren't the key — srchvrs · 2026-10-27
- First-week PhD student lands ARR Area Chair role, exposing peer-review quirks — anmarasovic · 2026-10-27