Research map / Large language models / Agents, code and math
AtomWorld: A Benchmark for Evaluating Spatial Reasoning in Large Language Models on Material Structures
Taoyuze Lv, Alexander Chen, Fengyu Xie, Chu Wu, Jeffrey Meng, Dongzhan Zhou, Bram Hoex, Yingheng Wang, Zhicheng Zhong, Tong Xie
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Pymatgen tool-augmented coding agent (ablation) | DeepSeek Chat · AtomMotor-2K (rotate around) | Success rate: 6.8% -> 18% | Improved |
| increase Qwen3 parameter count (4B to 32B) (ablation) | Qwen3 · AtomMotor-2K | Success rate | Improved |
| scale parameters on harder tasks (ablation) | Qwen3 · AtomMotor-2K (difficult actions) | Success rate | No clear effect |
| increase Qwen3 size (ablation) | Qwen3 · CCS | CCS score: 0.768 -> 1.141 | Improved |
| Qwen3-32B vs larger Llama3-70B | Qwen3-32B · CCS | CCS score: 0.987 (Llama3-70B) -> 1.141 | Improved |
| more atoms in structure (longer token length) (ablation) | LLMs · AtomMotor-2K | Success rate | Hurt |
| use Bravais lattice type or action position as varied condition (ablation) | LLMs · AtomMotor-2K | Success rate | No clear effect |
| use rotate around instead of move/insert between on bare coordinates (ablation) | Gemini 2.5 Pro, DeepSeek Chat · Pure coordinate test | Mean max dist: 0.000-0.005 -> 16.17 (Gemini) | Hurt |
| domain-specialized chemistry LLMs (NatureLM, LLaMat) | NatureLM, LLaMat · AtomMotor-2K (first 16 tasks per category) | Success rate: 0 | Hurt |
Closest papers by meaning
- Fine-Tuned Language Models Generate Stable Inorganic Materials as Text (ICLR 2024 · similarity 0.89)
- MatExpert: Decomposing Materials Discovery By Mimicking Human Experts (ICLR 2025 · similarity 0.88)
- FlowLLM: Flow Matching for Material Generation with Large Language Models as Base Distributions (NeurIPS 2024 · similarity 0.87)
- MADE: Benchmark Environments for Closed-Loop Materials Discovery (ICML 2026 · similarity 0.87)
- Scaling-Aware Adapter for Structure-Grounded LLM Reasoning (ICML 2026 · similarity 0.87)
- LLM Meets Diffusion: A Hybrid Framework for Crystal Material Generation (NeurIPS 2025 · similarity 0.87)
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.
