Graph-Based Exploration for ARC-AGI-3 Interactive Reasoning Tasks

TL;DR

ARC-AGI-3 challenges agents to discover game mechanics through interactive, sparse-feedback tasks. The paper presents a training-free graph-based framework that couples a Frame Processor for visual feature extraction with a Level Graph Explorer for systematic state-space exploration, using state hashing, priority-based action grouping, and frontier-driven traversal. Empirical results show the approach outperforms state-of-the-art LLM-based methods, solving a substantial number of levels across six games and achieving a top-three private leaderboard rank, albeit with a known bug affecting exact reset handling. The work demonstrates that explicit graph-structured exploration, guided by visual salience and frontier management, provides a strong foundation for interactive reasoning in environments where current LLMs struggle, and opens avenues for hybrids that combine graph-based exploration with learning-based models under sparse rewards.

Abstract

We present a training-free graph-based approach for solving interactive reasoning tasks in the ARC-AGI-3 benchmark. ARC-AGI-3 comprises game-like tasks where agents must infer task mechanics through limited interactions, and adapt to increasing complexity as levels progress. Success requires forming hypotheses, testing them, and tracking discovered mechanics. The benchmark has revealed that state-of-the-art LLMs are currently incapable of reliably solving these tasks. Our method combines vision-based frame processing with systematic state-space exploration using graph-structured representations. It segments visual frames into meaningful components, prioritizes actions based on visual salience, and maintains a directed graph of explored states and transitions. By tracking visited states and tested actions, the agent prioritizes actions that provide the shortest path to untested state-action pairs. On the ARC-AGI-3 Preview Challenge, this structured exploration strategy solves a median of 30 out of 52 levels across six games and ranks 3rd on the private leaderboard, substantially outperforming frontier LLM-based agents. These results demonstrate that explicit graph-structured exploration, even without learning, can serve as a strong baseline for interactive reasoning and underscore the importance of systematic state tracking and action prioritization in sparse-feedback environments where current LLMs fail to capture task dynamics. The code is open source and available at https://github.com/dolphin-in-a-coma/arc-agi-3-just-explore.

Figures (3)

• Figure 1: Effect of progressively adding method components to a random agent, compared with the LLM+DSL baseline. For each configuration, the stacked bar above the horizontal axis shows the total number of solved levels across the three private games, and the stacked bar below shows the total across the three public games. Colors indicate how many levels are solved in each individual game. The rightmost bars correspond to the full method. All non-LLM configurations report the median over 5 runs, whereas the LLM+DSL baseline is shown as a single result taken from the official challenge evaluation.
• Figure 2: Top row (Public set): vc33, ls20, ft09. Bottom row (Private set): sp80, lp85, as66.
• Figure 3: Levels solved as a function of environment steps for four methods: Random Agent, Random + Frame Segmentation, Random + Segmentation + New-Action Prioritization, and the full Graph Exploration method. The x-axis is logarithmic; each line shows the median over 5 runs and the shaded region shows the minimum–maximum range. Intermediate variants are shown up to 10,000 environment steps, while the Graph Explorer is plotted over the full evaluation budget.