我们评估 ARC-AGI-3 的初始编码代理系统，其中代理维护一个可执行的 Python 世界模型, 根据之前的观察结果验证它, 将其重构为更简单的抽象，作为类似 MDL 的简单性偏差的实际代理, 并在行动之前通过模型进行计划。该系统有意直接:它使用脚本控制器,预定义的世界模型接口,验证程序,和计划执行器,但没有手工编码的游戏特定逻辑。面向代理的提示, 工作区, 和控制器不包含特定于游戏的代码, 特定于游戏的提示, 手工编码启发式, 隐藏解决方案, 或其他特定于游戏的信息; 跨游戏使用相同的代理和提示。由于编码代理具有广泛的系统访问权限,，我们审核非预期信息通道, 描述早期的易受攻击的工具, 并解释当前工具如何关闭观察到的泄漏通道，同时减少特定于基准的信息暴露。我们报告 25 场公开 ARC-AGI-3 游戏的结果。每个游戏都从新的代理实例和干净的工作区, 开始，无法访问早期游戏中的文件或对话状态。凭借 GPT-5.5 高推理能力,，智能体完全解决了 15 场比赛，并实现了平均每场比赛 RHAE 为 58.12%。凭借 GPT-5.4 高推理努力,，它完全解决了 8 场比赛，并实现了平均每场比赛 RHAE 41.29%。我们尚未提供, 的私有验证集, 上的性能仍有待测试。总体而言, 结果提供了初步证据，证明验证者驱动的可执行世界模型对于 ARC-AGI-3 代理来说是一种有前途的方法。完整运行的工件与此 https URL 处的代码一起发布。
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. The agent-facing prompts, workspace, and controller contain no game-specific code, game-specific prompts, hand-coded heuristics, hidden solutions, or other game-specific information; the same agent and prompts are used across games. Because the coding agent has broad system access, we audit unintended information channels, describe earlier vulnerable harnesses, and explain how the current harness closes observed leakage channels while reducing benchmark-specific information exposure. We report results on the 25 public ARC-AGI-3 games. Each playthrough starts from a fresh agent instance and clean workspace, with no access to files or conversation state from earlier playthroughs. With GPT-5.5 high reasoning effort, the agent fully solved 15 games and achieved a mean per-game RHAE of 58.12%. With GPT-5.4 high reasoning effort, it fully solved 8 games and achieved a mean per-game RHAE of 41.29%. Performance on the private validation set, which is not yet available to us, 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. Full run artifacts are released with the code at this https URL.
科目:人工智能(cs.AI)
Subjects: Artificial Intelligence (cs.AI)