Schema agent harness hits 99% RHAE on ARC-AGI-3, paper and code released
xiuyu_l · x · 2026-10-13
- The team released Schema on July 15, the first agent harness to reach 99% RHAE on the ARC-AGI-3 public set: 99% with Opus 4.8 + Fable 5, and 95.35% with GPT-5.6 Sol.
- Today they published the full paper and code, with new results and a closer analysis of why Schema works.
- The core idea: make the LLM "think like a physicist," steering the model to build environment models at the harness level rather than relying on raw model capability alone.
More from coding & agent
- Yacine: all software is trash, so with AI it's now reasonable to just build it yourself — yacineMTB · 2026-10-13
- Claude Code spends 22 minutes and 120k tokens trying to add screws to screw holes — _Stocko_ · 2026-10-13
- NTU's ExpActivator boosts GUI agent step success 28% by activating only applicable experience — NanyangTechnologicalUniversity · 2026-10-13
- Dev builds agent sandboxing prototype with Pydantic and Tensorlake — samuelcolvin · 2026-10-13
- One dev rebuilt Battlerite in days with AI — playable in the browser — mattshumer_ · 2026-10-13
- Peter Steinberger's agentic framework quietly powers major new products — signulll · 2026-10-13