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.
More from coding & agent
- Grok Bot Introduces Zero-Barrier Coding Agent with No Model Selector — ccerrato147 · 2026-08-13
- Boosting Dev Workflow in the Agent Era with Xcode Ad-Hoc Builds — rounak · 2026-08-13
- ExcaliDash: A Self-Hosted Dashboard for Excalidraw with Live Collaboration — tom_doerr · 2026-08-13
- Omarchy Leans Fully Into the Age of AI Agents to Democratize Linux — dosco · 2026-08-13
- Qwen-CUA Launches Native Computer-Use Agent with Red Team Safety Benchmark — hhsun1 · 2026-08-13
- Qwen3.8-2.4T-A95B for long-running agents: 256K context, self-testing, function calling — togethercompute · 2026-08-13