"Today, we’re introducing [schema]: a harness reaching 99% RHAE with Opus 4.8 + Fable 5 and 95.35% with GPT-5.6 Sol on ARC-AGI-3 Public set. [schema] makes an LLM think like a physicist."
Today, we’re introducing [schema]: a harness reaching 99% RHAE with Opus 4.8 + Fable 5 and 95.35% with GPT-5.6 Sol on ARC-AGI-3 Public set.
[schema] makes an LLM think like a physicist. ARC-AGI-3 gives an agent a 64×64 grid plus legal actions, no rules, stated goal, or reward. The agent must discover both what the world is and how it works like a physicist:
- State grounding -> identify objects, relations, and goals.
- Mechanism discovery -> infer how these [schema] handles the state and mechanism in one editable program, a symbolic world model. It designs experiments to verify hypotheses, backtests the program against history, and plans inside its world at zero action cost. [schema]'s saturation of the ARC-AGI-3 public set is only a starting point. There is much more to explore!
Full blog: http:// schema-harness.github.io Agent traces: http:// huggingface.co/datasets/schem a-harness/arc-agi-3-schema-traces …
Amazing team effort with @guanningzeng , @JianiWang , @wenjie_ma , @shaofeng_y27736 , — Haven Feng