Claude Code Hits 96.2% on ARC-AGI-3 with Almost Zero ARC-Specific Code
khademinori · x · 2026-08-13
Developer @jerber used Claude Code paired with Opus 5 (high reasoning mode) to score 96.2% on the 25 public ARC-AGI-3 games, missing only one. A second attempt (pass@2) achieved 99.3%, costing about $540 in total.
For comparison, Opus 5 scores just 30.2% in the official model-only evaluation. The core highlight of this approach is that it uses almost zero ARC-specific code. The harness consists merely of a shell, a filesystem, and a single command for taking actions—there is no pre-built solver, planner, or grid tooling.
Instead, the model writes and discards these tools on the fly while playing. In a single pass, it generated 269 programs (about 12,700 lines of code, ten times the size of the harness itself) and discarded all of them once their games ended. The author has open-sourced the harness and several complete session records.
Related event: Opus 5 Tops ARC-AGI-3 via Claude Code(7 posts)→
More from coding & agent
- Dev builds MCP server that lets AI agents outsource phone calls to real humans — Many_Benefit_3393 · 2026-10-16
- A vector DB is not LLM memory: long thread maps write/read paths, failure modes and evals — techNmak · 2026-10-16
- Mergelock: open-source MCP server catches semantic merge conflicts git misses across parallel AI agents — ksadayamuthu · 2026-10-16
- OpenUI 1.0 launches open source, turning agent text answers into live interactive UIs — heyshrutimishra · 2026-10-16
- 10 GitHub repos to see what your AI agent really did, from Langfuse to claude-code-log — victor_explore · 2026-10-16
- Mosaic: multi-session agentic RL with hierarchical advantage estimation — LARK-Lab · 2026-10-16