{"id":"W4417402430","doi":"10.2196/77978","title":"Multiple Large Language Models’ Performance on the Chinese Medical Licensing Examination: Quantitative Comparative Study","year":2025,"lang":"en","type":"article","venue":"JMIR Human Factors","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Medical practice; MEDLINE; Clinical Practice; Public health; Medical knowledge","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01300249,0.001592689,0.0009777891,0.001942969,0.0007714627,0.002135738,0.001479848,0.001263609,0.001881144],"category_scores_gemma":[0.03049724,0.0003955459,0.001165797,0.00131681,0.0007345466,0.002930348,0.002056068,0.001671942,0.001202757],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002716035,"about_ca_system_score_gemma":0.002584453,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.03122933,"about_ca_topic_score_gemma":0.02515579,"domain_scores_codex":[0.994222,0.003171043,0.0004161143,0.00101417,0.0008640597,0.0003126822],"domain_scores_gemma":[0.9771533,0.01701105,0.0008470253,0.001308573,0.002693213,0.0009868219],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.003498992,0.002833125,0.2799733,0.001602622,0.001772656,0.0007314727,0.003419361,0.1525958,0.007086912,0.001629158,0.02351202,0.5213446],"study_design_scores_gemma":[0.0002207462,0.001508758,0.06917458,0.000137588,0.0007721079,0.000208921,0.001811105,0.9142932,0.005807409,0.001373308,0.004540278,0.000151971],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9784104,0.002290572,0.009532658,0.001016593,0.000156952,0.0002333643,0.001708338,0.001888616,0.004762426],"genre_scores_gemma":[0.984866,0.0003650802,0.009530211,0.0001533858,0.00005983788,0.0001491294,0.003507286,0.00009820427,0.001270831],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03122933,"threshold_uncertainty_score":0.06876463,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2278652237110262,"score_gpt":0.4957672118267702,"score_spread":0.2679019881157439,"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."}}