{"id":"W4410022944","doi":"10.1088/2632-2153/add3bb","title":"LLM4Mat-bench: benchmarking large language models for materials property prediction","year":2025,"lang":"en","type":"article","venue":"Machine Learning Science and Technology","topic":"Machine Learning in Materials Science","field":"Materials Science","cited_by":13,"is_retracted":false,"has_abstract":true,"ca_institutions":"Schwartz/Reisman Emergency Medicine Institute; Vector Institute; Structural Genomics Consortium; University of Toronto","funders":"Office of Advanced Cyberinfrastructure; National Science Foundation","keywords":"Benchmarking; Property (philosophy); Computer science; Natural language processing; Business; Philosophy","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00443031,0.003976991,0.001397726,0.002084561,0.001137997,0.002544394,0.006468853,0.003731416,0.01464288],"category_scores_gemma":[0.01814738,0.001139148,0.002796454,0.002027385,0.0012165,0.004352475,0.002868287,0.003578939,0.01140755],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002406091,"about_ca_system_score_gemma":0.002898069,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01839255,"about_ca_topic_score_gemma":0.02317271,"domain_scores_codex":[0.9965689,0.0012764,0.0002685078,0.000946487,0.0006760317,0.0002637716],"domain_scores_gemma":[0.9923906,0.004911747,0.0002361016,0.00124115,0.0008898721,0.0003305133],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.002387342,0.001561955,0.01076982,0.005636088,0.001309119,0.0008292776,0.0004645703,0.3340397,0.01106269,0.01024192,0.4374133,0.1842842],"study_design_scores_gemma":[0.0003592769,0.000345524,0.001538785,0.0001354841,0.00009381147,0.0001268927,0.00017351,0.9468352,0.01319444,0.008350138,0.02873861,0.0001083717],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"software","genre_gemma":"empirical","genre_scores_codex":[0.1923108,0.009860672,0.1474311,0.004259504,0.00246583,0.001046577,0.1568757,0.465267,0.02048286],"genre_scores_gemma":[0.3904717,0.001440963,0.1985486,0.002229654,0.0002523107,0.001526239,0.3813144,0.01587138,0.008344878],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01839255,"threshold_uncertainty_score":0.04898524,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.007163373249796719,"score_gpt":0.2604273323180107,"score_spread":0.253263959068214,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}