Executable World Models for ARC-AGI-3 in the Era of Coding Agents

TL;DR

This paper presents a coding-agent system with an executable Python world model that verifies, refactors, and plans before acting, demonstrating promising results on ARC-AGI-3 games without game-specific code.

Contribution

It introduces a verifier-driven executable world model approach for ARC-AGI-3, serving as a game-general baseline with no hand-coded game logic.

Findings

Solved 7 out of 25 ARC-AGI-3 games.

Achieved over 75% relative human action efficiency on 6 games.

Mean per-game RHAE of 32.58%.

Abstract

We evaluate an initial coding-agent system for ARC-AGI-3 in which the agent maintains an executable Python world model, verifies it against previous observations, refactors it toward simpler abstractions as a practical proxy for an MDL-like simplicity bias, and plans through the model before acting. The system is intentionally direct: it uses a scripted controller, predefined world-model interfaces, verifier programs, and a plan executor, but no hand-coded game-specific logic. We report results on the 25 public ARC-AGI-3 games. Each recorded playthrough uses a fresh agent instance with no access to previous playthrough-specific files or conversation state. Most games have a single recorded playthrough; for a few games, we report multiple independent fresh-agent playthroughs to expose run-to-run variability. The agent fully solved 7 games, achieved a Relative Human Action Efficiency greater than 75%, on 6 games, and obtained a mean per-game RHAE of 32.58%. Because the system uses no game-specific code, it can serve as a game-general baseline for ARC-AGI-3. Performance on the private validation set remains to be tested. Overall, the results provide preliminary evidence that verifier-driven executable world models are a promising approach for ARC-AGI-3 agents.