AI Paper Atlas

Research map / Multi-agent systems and games / LLM agents

CubeBench: Diagnosing Interactive, Long-Horizon Physical Intelligence under Partial Observations

Huan-ang Gao, Zikang Zhang, Tianwei Luo, Kaisen Yang, Xinzhe Juan, Jiahao Qiu, Tianxing Chen, Bingxiang He, Hao Zhao, Hao Zhou, Shilong Liu, Mengdi Wang

ICLR 2026 · Multi-agent systems and games › LLM agents · approach: Agentic benchmarks and orchestration

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Introduces CubeBench, a benchmark for diagnosing LLM agents' physical intelligence on long-horizon, partially observed interactive tasks, targeting spatial reasoning, state tracking via mental simulation, and active exploration.

Problem

LLM agents struggle to transfer from digital tasks to the physical world because they must form and maintain a robust spatial mental model under partial observation over long horizons.

Approach

Isolates three cognitive challenges (spatial reasoning, long-horizon state tracking via mental simulation, active exploration under partial observation) in a diagnostic benchmark so each can be evaluated separately.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add dense face reward to Basic Agent (ablation)GPT-5 · CubeBench Full Symbolic short-horizonpass rate: 0.75 -> 0.85Improved
add dense sticker or heuristic reward to Basic Agent (ablation)GPT-5 · CubeBench Full Symbolic short-horizonpass rate: 0.75 -> 0.65 (sticker), 0.50 (heuristic)Hurt
add dense rewards (any type) on long-horizon tasks (ablation)all models · CubeBench long-horizonpass rate: 0.00 -> 0.00No clear effect
add dense sticker reward (ablation)Claude Sonnet 4 · CubeBench Full Symbolic short-horizonpass rate: 0.05 -> 0.25Improved
add dense face reward to Basic Agent (ablation)GPT-5 · CubeBench Full Visual short-horizonpass rate: 0.20 -> 0.55Improved
add Standard-Solver tool (ablation)GPT-5 · CubeBench Full Symbolic long-horizonpass rate: 0.00 -> 0.95Improved
add Standard-Solver tool (ablation)Claude Sonnet 4 · CubeBench Full Symbolic long-horizonpass rate: 0.00 -> 0.85Improved
replace Standard-Solver with Ideal-Solver (auto format conversion) (ablation)Gemini 2.5 Pro · CubeBench Full Symbolic short-horizonpass rate: 0.70 -> 1.00Improved
replace Standard-Solver with Ideal-Solver (ablation)Claude Sonnet 4 · CubeBench Full Visualpass rate: 0.00 -> 0.00No clear effect
add solver tools on Vertex view (ablation)all models · CubeBench Vertex viewpass rate: 0.00 -> 0.00No clear effect
remove code execution (No-Code mode) (ablation)GPT-5 · CubeBench Full Symbolic short-horizonpass rate: 0.75 -> 0.25Hurt
remove code execution (No-Code mode) (ablation)Grok-4 · CubeBench Full Symbolic short-horizonpass rate: 0.20 -> 0.30Improved

Experimental setup

Benchmarks: CubeBench

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy