CubeBench: Diagnosing Interactive, Long-Horizon Spatial Reasoning Under Partial Observations

TL;DR

This paper introduces CubeBench, a benchmark for evaluating LLM agents' spatial reasoning and long-horizon planning in physical tasks using Rubik's Cube, revealing significant limitations in current models.

Contribution

It presents CubeBench, a novel diagnostic benchmark with a three-tiered framework to assess spatial reasoning, state tracking, and exploration in LLMs for physical tasks.

Findings

Leading LLMs fail all long-horizon tasks with 0% success

Identifies key cognitive bottlenecks in spatial reasoning and planning

Provides insights for developing more physically-grounded agents

Abstract

Large Language Model (LLM) agents, while proficient in the digital realm, face a significant gap in physical-world deployment due to the challenge of forming and maintaining a robust spatial mental model. We identify three core cognitive challenges hindering this transition: spatial reasoning, long-horizon state tracking via mental simulation, and active exploration under partial observation. To isolate and evaluate these faculties, we introduce CubeBench, a novel generative benchmark centered on the Rubik's Cube. CubeBench uses a three-tiered diagnostic framework that progressively assesses agent capabilities, from foundational state tracking with full symbolic information to active exploration with only partial visual data. Our experiments on leading LLMs reveal critical limitations, including a uniform 0.00% pass rate on all long-horizon tasks, exposing a fundamental failure in long-term planning. We also propose a diagnostic framework to isolate these cognitive bottlenecks by providing external solver tools. By analyzing the failure modes, we provide key insights to guide the development of more physically-grounded intelligent agents.