{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,10,21]],"date-time":"2026-10-21T22:39:55Z","timestamp":1792622395681,"version":"4.5.0"},"reference-count":44,"publisher":"Springer Science and Business Media LLC","issue":"10","license":[{"start":{"date-parts":[[2026,10,12]],"date-time":"2026-10-12T00:00:00Z","timestamp":1791763200000},"content-version":"tdm","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0"},{"start":{"date-parts":[[2026,10,12]],"date-time":"2026-10-12T00:00:00Z","timestamp":1791763200000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0"}],"funder":[{"DOI":"10.13039\/100007569","name":"Carl-Zeiss-Stiftung","doi-asserted-by":"publisher","id":[{"id":"10.13039\/100007569","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/501100009318","name":"Helmholtz Association","doi-asserted-by":"publisher","id":[{"id":"10.13039\/501100009318","id-type":"DOI","asserted-by":"publisher"}]},{"name":"Intel and Merck via the AWASES research center"}],"content-domain":{"domain":["link.springer.com"],"crossmark-restriction":false},"short-container-title":["Nat Comput Sci"],"abstract":"<jats:title>Abstract<\/jats:title>\n                  <jats:p>\n                    Predicting properties from coordinate\u2013category data\u2014sets of vectors paired with categorical information\u2014is fundamental to computational science, manifesting in materials science as predicting formation energies or elastic moduli from crystal structures. While large language models (LLMs) have increasingly been applied to such tasks, optimal strategies for achieving reliable predictions remain elusive. Here we report fundamental limitations in LLM\u2019s ability to learn from coordinate information in coordinate\u2013category data. Through systematic experiments using synthetic datasets with tunable coordinate and category contributions, and a comprehensive benchmarking framework (MatText), we find that LLMs consistently fail to capture coordinate information while excelling at category patterns. This geometric blindness persists regardless of model size (up to 70 billion parameters), dataset scale (up to 2 \u00d7 10\n                    <jats:sup>6<\/jats:sup>\n                    structures) or text representation strategy. For material property prediction tasks dominated by structural effects, specialized geometric architectures consistently outperform LLMs, as evidenced by a clear \u2018Graph Neural Network Language Model (GNN-LM) wall\u2019 in performance benchmarks. Our analysis provides concrete guidelines for architecture selection in scientific machine learning.\n                  <\/jats:p>","DOI":"10.1038\/s43588-026-01045-2","type":"journal-article","created":{"date-parts":[[2026,10,12]],"date-time":"2026-10-12T15:54:28Z","timestamp":1791820468000},"page":"1124-1132","update-policy":"https:\/\/doi.org\/10.1007\/springer_crossmark_policy","source":"Crossref","is-referenced-by-count":0,"title":["Less can be more for predicting properties with large language models"],"prefix":"10.1038","volume":"6","author":[{"ORCID":"https:\/\/orcid.org\/0009-0001-7697-7315","authenticated-orcid":false,"given":"Nawaf","family":"Alampara","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Santiago","family":"Miret","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-4894-4660","authenticated-orcid":false,"given":"Kevin Maik","family":"Jablonka","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"297","published-online":{"date-parts":[[2026,10,12]]},"reference":[{"key":"1045_CR1","doi-asserted-by":"publisher","unstructured":"Bommasani, R. et al. On the opportunities and risks of foundation models. Preprint at https:\/\/doi.org\/10.48550\/arXiv.2108.07258 (2021).","DOI":"10.48550\/arXiv.2108.07258"},{"key":"1045_CR2","doi-asserted-by":"publisher","first-page":"2484","DOI":"10.1021\/acs.chemrev.5c00583","volume":"126","author":"N Alampara","year":"2026","unstructured":"Alampara, N. et al. General-purpose models for the chemical sciences: LLMs and beyond. Chem. Rev. 126, 2484\u20132549 (2026).","journal-title":"Chem. Rev."},{"key":"1045_CR3","doi-asserted-by":"publisher","first-page":"1027","DOI":"10.1038\/s41557-025-01815-x","volume":"17","author":"A Mirza","year":"2025","unstructured":"Mirza, A. et al. A framework for evaluating the chemical knowledge and reasoning abilities of large language models against the expertise of chemists. Nat. Chem. 17, 1027\u20131034 https:\/\/doi.org\/10.1038\/s41557-025-01815-x (2025).","journal-title":"Nat. Chem."},{"key":"1045_CR4","doi-asserted-by":"publisher","first-page":"952","DOI":"10.1038\/s43588-025-00836-3","volume":"5","author":"N Alampara","year":"2025","unstructured":"Alampara, N. et al. Probing the limitations of multimodal language models for chemistry and materials research. Nat. Comput. Sci. 5, 952\u2013961 (2025).","journal-title":"Nat. Comput. Sci."},{"key":"1045_CR5","doi-asserted-by":"publisher","first-page":"161","DOI":"10.1038\/s42256-023-00788-1","volume":"6","author":"KM Jablonka","year":"2024","unstructured":"Jablonka, K. M., Schwaller, P., Ortega-Guerrero, A. & Smit, B. Leveraging large language models for predictive chemistry. Nat. Mach. Intell. 6, 161\u2013169 https:\/\/doi.org\/10.1038\/s42256-023-00788-1 (2024).","journal-title":"Nat. Mach. Intell."},{"key":"1045_CR6","doi-asserted-by":"publisher","DOI":"10.1038\/s41524-025-01536-2","volume":"11","author":"A Niyongabo Rubungo","year":"2025","unstructured":"Niyongabo Rubungo, A., Arnold, C., Rand, B. P. & Dieng, A. B. LLM-Prop: predicting the properties of crystalline materials using large language models. npj Comput. Mater. 11, 186 (2025).","journal-title":"npj Comput. Mater."},{"key":"1045_CR7","unstructured":"Berglund, L. et al. The reversal curse: LLMs trained on \"A is B\" fail to learn \"B is A\". In 13th International Conference on Learning Representations (eds Kim, B. et al.) 18623\u201318642 (Curran Assoc., 2024)"},{"key":"1045_CR8","unstructured":"Mirzadeh, I. et al. GSM-Symbolic: understanding the limitations of mathematical reasoning in large language models. In 13th International Conference on Learning Representations (eds Yue, Y. et al.) 94743\u201394765 (Curran, Assoc., 2025)."},{"key":"1045_CR9","doi-asserted-by":"publisher","first-page":"991","DOI":"10.1038\/s42256-025-01058-y","volume":"7","author":"S Miret","year":"2025","unstructured":"Miret, S. & Krishnan, N. M. A. Enabling large language models for real-world materials discovery. Nat. Mach. Intell. 7, 991\u2013998 (2025).","journal-title":"Nat. Mach. Intell."},{"key":"1045_CR10","doi-asserted-by":"publisher","unstructured":"R\u00edos-Garc\u00eda, M., Alampara, N. & Jablonka, K. M. Semantic content determines algorithmic performance. Preprint at https:\/\/doi.org\/10.48550\/arXiv.2601.21618 (2026).","DOI":"10.48550\/arXiv.2601.21618"},{"key":"1045_CR11","doi-asserted-by":"publisher","unstructured":"Reiser, P. et al. Graph neural networks for materials science and chemistry. Commun. Mater. 3, 93 https:\/\/doi.org\/10.1038\/s43246-022-00315-6 (2022).","DOI":"10.1038\/s43246-022-00315-6"},{"key":"1045_CR12","doi-asserted-by":"publisher","unstructured":"Choudhary, K. et al. Recent advances and applications of deep learning methods in materials science. npj Comput. Mater. 8, 59 https:\/\/doi.org\/10.1038\/s41524-022-00734-6 (2022).","DOI":"10.1038\/s41524-022-00734-6"},{"key":"1045_CR13","unstructured":"Gruver, N. et al. Fine-tuned language models generate stable inorganic materials as text. In 13th International Conference on Learning Representations (eds Kim, B. et al.) 21671\u201321690 (Curran Assoc., 2024)."},{"key":"1045_CR14","doi-asserted-by":"crossref","unstructured":"Devlin, J. et al. BERT: pre-training of deep bidirectional transformers for language understanding. In Proc. 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) (eds Burstein, J. et al.) 4171\u20134186 (ACL, 2019).","DOI":"10.18653\/v1\/N19-1423"},{"key":"1045_CR15","doi-asserted-by":"publisher","first-page":"594","DOI":"10.1039\/D4DD00018H","volume":"3","author":"R Ruff","year":"2024","unstructured":"Ruff, R., Reiser, P., St\u00fchmer, J. & Friederich, P. Connectivity optimized nested line graph networks for crystal structures. Digit. Discov. 3, 594\u2013601 (2024).","journal-title":"Digit. Discov."},{"key":"1045_CR16","doi-asserted-by":"publisher","first-page":"404002","DOI":"10.1088\/1361-648X\/ac1280","volume":"33","author":"P-P De Breuck","year":"2021","unstructured":"De Breuck, P.-P., Evans, M. L. & Rignanese, G.-M. Robust model benchmarking and bias-imbalance in data-driven materials science: a case study on modnet. J. Phys. Condens. Matter 33, 404002 (2021).","journal-title":"J. Phys. Condens. Matter"},{"key":"1045_CR17","doi-asserted-by":"publisher","first-page":"98049","DOI":"10.52202\/079017-3112","volume":"37","author":"T Nguyen","year":"2024","unstructured":"Nguyen, T. Understanding transformers via n-gram statistics. Adv. Neural Inf. Process. Syst. 37, 98049\u201398082 (2024).","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"1045_CR18","unstructured":"Zhou, H. et al. What algorithms can transformers learn? A study in length generalization. In 13th International Conference on Learning Representations (eds Kim, B. et al.) 15898\u201315926 (Curran Assoc., 2024)."},{"key":"1045_CR19","doi-asserted-by":"publisher","DOI":"10.1038\/s41467-023-42870-7","volume":"14","author":"H Xiao","year":"2023","unstructured":"Xiao, H. et al. An invertible, invariant crystal representation for inverse design of solid-state materials using generative deep learning. Nat. Commun. 14, 7027 (2023).","journal-title":"Nat. Commun."},{"key":"1045_CR20","doi-asserted-by":"publisher","unstructured":"Kaplan, J. et al. Scaling laws for neural language models. Preprint at https:\/\/doi.org\/10.48550\/arXiv.2001.08361 (2020).","DOI":"10.48550\/arXiv.2001.08361"},{"key":"1045_CR21","doi-asserted-by":"crossref","unstructured":"Hoffmann, J. et al. An empirical analysis of compute-optimal large language model training. In Advances in Neural Information Processing Systems (eds Koyejo, S. et al.) 30016\u201330030 (NeurIPS, 2022).","DOI":"10.52202\/068431-2176"},{"key":"1045_CR22","doi-asserted-by":"publisher","first-page":"1297","DOI":"10.1038\/s42256-023-00740-3","volume":"5","author":"NC Frey","year":"2023","unstructured":"Frey, N. C. et al. Neural scaling of deep chemical models. Nat. Mach. Intell. 5, 1297\u20131305 (2023).","journal-title":"Nat. Mach. Intell."},{"key":"1045_CR23","doi-asserted-by":"publisher","unstructured":"McKenzie, I. R. et al. Inverse scaling: when bigger isn\u2019t better. Preprint at https:\/\/doi.org\/10.48550\/arXiv.2306.09479 (2023).","DOI":"10.48550\/arXiv.2306.09479"},{"key":"1045_CR24","doi-asserted-by":"publisher","unstructured":"Dunn, A., Wang, Q., Ganose, A., Dopp, D. & Jain, A. Benchmarking materials property prediction methods: the matbench test set and automatminer reference algorithm. npj Comput. Mater. 6, 138 https:\/\/doi.org\/10.1038\/s41524-020-00406-3 (2020).","DOI":"10.1038\/s41524-020-00406-3"},{"key":"1045_CR25","doi-asserted-by":"publisher","first-page":"13890","DOI":"10.1038\/ncomms13890","volume":"8","author":"KT Sch\u00fctt","year":"2017","unstructured":"Sch\u00fctt, K. T., Arbabzadah, F., Chmiela, S., M\u00fcller, K.-R. & Tkatchenko, A. Quantum-chemical insights from deep tensor neural networks. Nat. Commun. 8, 13890 (2017).","journal-title":"Nat. Commun."},{"key":"1045_CR26","doi-asserted-by":"publisher","unstructured":"Gasteiger, J., Gro\u00df, J. & G\u00fcnnemann, S. Directional message passing for molecular graphs. Preprint at https:\/\/doi.org\/10.48550\/arXiv.2003.03123 (2020).","DOI":"10.48550\/arXiv.2003.03123"},{"key":"1045_CR27","doi-asserted-by":"publisher","DOI":"10.1038\/s41524-021-00545-1","volume":"7","author":"AY-T Wang","year":"2021","unstructured":"Wang, A. Y.-T., Kauwe, S. K., Murdock, R. J. & Sparks, T. D. Compositionally restricted attention-based network for materials property predictions. npj Comput. Mater. 7, 77 (2021).","journal-title":"npj Comput. Mater."},{"key":"1045_CR28","doi-asserted-by":"publisher","DOI":"10.1038\/s41524-024-01231-8","volume":"10","author":"J Qu","year":"2024","unstructured":"Qu, J. et al. Leveraging language representation for materials exploration and discovery. npj Comput. Mater. 10, 58 (2024).","journal-title":"npj Comput. Mater."},{"key":"1045_CR29","doi-asserted-by":"publisher","first-page":"10024","DOI":"10.1021\/ja00051a040","volume":"114","author":"AK Rapp\u00e9","year":"1992","unstructured":"Rapp\u00e9, A. K., Casewit, C. J., Colwell, K. S., Goddard III, W. A. & Skiff, W. M. UFF, a full periodic table force field for molecular mechanics and molecular dynamics simulations. J. Am. Chem. Soc. 114, 10024\u201310035 (1992).","journal-title":"J. Am. Chem. Soc."},{"key":"1045_CR30","doi-asserted-by":"publisher","first-page":"493","DOI":"10.1108\/00220410410560573","volume":"60","author":"K Sp\u00e4rck Jones","year":"2004","unstructured":"Sp\u00e4rck Jones, K. A statistical interpretation of term specificity and its application in retrieval. J. Doc. 60, 493\u2013502 (2004).","journal-title":"J. Doc."},{"key":"1045_CR31","doi-asserted-by":"crossref","unstructured":"Chen, T. & Guestrin, C. XGBoost: a scalable tree boosting system. In Proc. 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (eds Krishnapuram, B. et al.) 785\u2013794 (ACM, 2016)","DOI":"10.1145\/2939672.2939785"},{"key":"1045_CR32","doi-asserted-by":"publisher","first-page":"874","DOI":"10.1557\/mrc.2019.94","volume":"9","author":"AM Ganose","year":"2019","unstructured":"Ganose, A. M. & Jain, A. Robocrystallographer: automated crystal structure text descriptions and analysis. MRS Commun. 9, 874\u2013881 (2019).","journal-title":"MRS Commun."},{"key":"1045_CR33","doi-asserted-by":"publisher","first-page":"6682","DOI":"10.1021\/acs.cgd.9b01050","volume":"19","author":"BJ Bucior","year":"2019","unstructured":"Bucior, B. J. et al. Identification schemes for metal-organic frameworks to enable rapid search and cheminformatics analysis. Cryst. Growth Des. 19, 6682\u20136697 (2019).","journal-title":"Cryst. Growth Des."},{"key":"1045_CR34","doi-asserted-by":"publisher","first-page":"5388","DOI":"10.21105\/joss.05388","volume":"8","author":"M Scheidgen","year":"2023","unstructured":"Scheidgen, M. et al. NOMAD: a distributed web-based platform for managing materials science research data. J. Open Source Softw. 8, 5388 (2023).","journal-title":"J. Open Source Softw."},{"key":"1045_CR35","doi-asserted-by":"publisher","unstructured":"Lee, K. L. K. et al. MatSciML: a broad, multi-task benchmark for solid-state materials modeling. Preprint at https:\/\/doi.org\/10.48550\/arXiv.2309.05934 (2023).","DOI":"10.48550\/arXiv.2309.05934"},{"key":"1045_CR36","doi-asserted-by":"publisher","first-page":"60","DOI":"10.1016\/j.commatsci.2018.05.018","volume":"152","author":"L Ward","year":"2018","unstructured":"Ward, L. et al. Matminer: an open source toolkit for materials data mining. Comput. Mater. Sci. 152, 60\u201369 (2018).","journal-title":"Comput. Mater. Sci."},{"key":"1045_CR37","doi-asserted-by":"publisher","first-page":"9034","DOI":"10.1039\/c2ee22341d","volume":"5","author":"IE Castelli","year":"2012","unstructured":"Castelli, I. E. et al. New cubic perovskites for one-and two-photon water splitting using the computational materials repository. Energy Environ. Sci. 5, 9034\u20139043 (2012).","journal-title":"Energy Environ. Sci."},{"key":"1045_CR38","doi-asserted-by":"publisher","first-page":"150009","DOI":"10.1038\/sdata.2015.9","volume":"2","author":"M de Jong","year":"2015","unstructured":"de Jong, M. et al. Charting the complete elastic properties of inorganic crystalline compounds. Sci. Data 2, 150009 (2015).","journal-title":"Sci. Data"},{"key":"1045_CR39","doi-asserted-by":"publisher","unstructured":"Touvron, H. et al. Llama 2: open foundation and fine-tuned chat models. Preprint at https:\/\/doi.org\/10.48550\/arXiv.2307.09288 (2023).","DOI":"10.48550\/arXiv.2307.09288"},{"key":"1045_CR40","doi-asserted-by":"publisher","unstructured":"Hu, E. J. et al. Lora: low-rank adaptation of large language models. Preprint at https:\/\/doi.org\/10.48550\/arXiv.2106.09685 (2021).","DOI":"10.48550\/arXiv.2106.09685"},{"key":"1045_CR41","doi-asserted-by":"crossref","unstructured":"Dettmers, T., Pagnoni, A., Holtzman, A. & Zettlemoyer, L. QLoRA: efficient finetuning of quantized LLMs. In Advances in Neural Information Processing Systems 36 (eds Oh, A. et al.) 10088\u201310115 (Curran Assoc., 2023).","DOI":"10.52202\/075280-0441"},{"key":"1045_CR42","doi-asserted-by":"crossref","unstructured":"Song, Y., Miret, S., Zhang, H. & Liu, B. HoneyBee: progressive instruction finetuning of large language models for materials science. In Findings of the Association for Computational Linguistics: EMNLP 2023 (eds Bouamor, H., Pino, J. & Bali, K.) 5724\u20135739 https:\/\/aclanthology.org\/2023.findings-emnlp.380\/ (ACL, 2023).","DOI":"10.18653\/v1\/2023.findings-emnlp.380"},{"key":"1045_CR43","doi-asserted-by":"publisher","unstructured":"Alampara, N. n0w0f\/MatText. HuggingFace https:\/\/doi.org\/10.57967\/hf\/2363 (2024).","DOI":"10.57967\/hf\/2363"},{"key":"1045_CR44","doi-asserted-by":"publisher","unstructured":"Alampara, N., Miret, S., & Jablonka, K. M. MatText: Less can be more for predicting properties with large language models (code) (Version 1.1.0). Zenodo. https:\/\/doi.org\/10.5281\/zenodo.21151839 (2026).","DOI":"10.5281\/zenodo.21151839"}],"container-title":["Nature Computational Science"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/www.nature.com\/articles\/s43588-026-01045-2.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/www.nature.com\/articles\/s43588-026-01045-2","content-type":"text\/html","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/www.nature.com\/articles\/s43588-026-01045-2.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,10,21]],"date-time":"2026-10-21T22:06:34Z","timestamp":1792620394000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.nature.com\/articles\/s43588-026-01045-2"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,10,12]]},"references-count":44,"journal-issue":{"issue":"10","published-online":{"date-parts":[[2026,10]]}},"alternative-id":["1045"],"URL":"https:\/\/doi.org\/10.1038\/s43588-026-01045-2","relation":{},"ISSN":["2662-8457"],"issn-type":[{"value":"2662-8457","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,10,12]]},"assertion":[{"value":"9 July 2025","order":1,"name":"received","label":"Received","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"7 August 2026","order":2,"name":"accepted","label":"Accepted","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"12 October 2026","order":3,"name":"first_online","label":"First Online","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"K.M.J. has been a paid contractor for OpenAI (as part of the red teaming network). The other authors declare no competing interests.","order":1,"name":"Ethics","label":"Competing interests","group":{"name":"EthicsHeading","label":"Ethics"}}]}}