{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,6,17]],"date-time":"2026-06-17T05:51:22Z","timestamp":1781675482198,"version":"3.54.5"},"reference-count":25,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2026,6,17]],"date-time":"2026-06-17T00:00:00Z","timestamp":1781654400000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":["frontiersin.org"],"crossmark-restriction":true},"short-container-title":["Front. Oncol."],"abstract":"<jats:sec>\n                    <jats:title>Introduction<\/jats:title>\n                    <jats:p>The rapid evolution of large language models has demonstrated human-level proficiency in general English-language medical examinations, yet their applicability in highly specialized, non-English clinical domains remains insufficiently explored.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Methods<\/jats:title>\n                    <jats:p>To bridge this gap, this study benchmarked leading domestic (Qwen 3 Max, DeepSeek V3.2) and international (GPT-5, Gemini 2.5 Flash) models against a board-certified Chinese radiation oncologist, using the National Intermediate Professional Title Examination for Radiation Oncology in China. Statistical comparisons were conducted using mixed-effects logistic regression and McNemar's test for paired data.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Results<\/jats:title>\n                    <jats:p>Domestic models demonstrated strong performance, with Qwen achieving an accuracy of 86.30%, which was higher than that of the single physician reference participant (adjusted P = 0.020). However, because only one human participant was included, this comparison should be interpreted as an illustrative reference rather than evidence of general superiority over radiation oncologists. In contrast, international models exhibited a marked performance decline in this specific context. Detailed cognitive stratification revealed a distinct reasoning-recall dissociation in international models, which maintained relatively robust clinical reasoning capabilities but failed significantly in localized knowledge retrieval. Notably, translating the examination into English did not bridge this performance gap for international models, while revealing a significant language penalty for certain domestic architectures (e.g., DeepSeek, P = 0.013). Granular error analysis confirmed that the majority of failures in international models like GPT-5 stemmed from discrepancies between Western and Chinese clinical guidelines rather than intrinsic reasoning deficits or hallucinations.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Discussion<\/jats:title>\n                    <jats:p>These findings suggest that, within this examination-based Chinese radiation oncology benchmark, alignment with regional clinical standards may be a major contributor to model performance. However, differences in model architecture, training data, post-training, interface configuration, and potential test-set contamination may also contribute.<\/jats:p>\n                  <\/jats:sec>","DOI":"10.3389\/fonc.2026.1808714","type":"journal-article","created":{"date-parts":[[2026,6,17]],"date-time":"2026-06-17T04:51:49Z","timestamp":1781671909000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":0,"title":["Knowledge localization is associated with higher performance of domestic large language models in a Chinese radiation oncology examination"],"prefix":"10.3389","volume":"16","author":[{"given":"Yuchen","family":"Zhou","sequence":"first","affiliation":[{"name":"Department of Radiation Oncology, Peking Union Medical College Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College","place":["Beijing, China"]},{"name":"Tsinghua Medicine, School of Medicine, Tsinghua University","place":["Beijing, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Shuyu","family":"Lin","sequence":"additional","affiliation":[{"name":"Department of Radiation Oncology, National Cancer Center\/National Clinical Research Center For Cancer\/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College","place":["Beijing, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Xinhai","family":"Wang","sequence":"additional","affiliation":[{"name":"Department of Radiation Oncology, Peking Union Medical College Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College","place":["Beijing, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Ke","family":"Hu","sequence":"additional","affiliation":[{"name":"Department of Radiation Oncology, Peking Union Medical College Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College","place":["Beijing, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1965","published-online":{"date-parts":[[2026,6,17]]},"reference":[{"key":"B1","doi-asserted-by":"publisher","first-page":"44","DOI":"10.1038\/s41591-018-0300-7","article-title":"High-performance medicine: the convergence of human and artificial intelligence","volume":"25","author":"Topol","year":"2019","journal-title":"Nat Med"},{"key":"B2","doi-asserted-by":"publisher","DOI":"10.1038\/s41591-023-02448-8","article-title":"Large language models in medicine","volume":"29","author":"Thirunavukarasu","year":"2023","journal-title":"Nat Med"},{"key":"B3","doi-asserted-by":"publisher","DOI":"10.1038\/s41586-023-06291-2","article-title":"Large language models encode clinical knowledge","volume":"620","author":"Singhal","year":"2023","journal-title":"Nature"},{"key":"B4","doi-asserted-by":"publisher","DOI":"10.1371\/journal.pdig.0000198","article-title":"Performance of ChatGPT on USMLE: potential for AI-assisted medical education using large language models","volume":"2","author":"Kung","year":"2023","journal-title":"PloS Digit Health"},{"key":"B5","doi-asserted-by":"publisher","DOI":"10.2196\/50442","article-title":"Assessment of ChatGPT-3.5\u2019s knowledge in oncology: comparative study with ASCO-SEP benchmarks","volume":"3","author":"Odabashian","year":"2024","journal-title":"JMIR AI"},{"key":"B6","doi-asserted-by":"publisher","DOI":"10.1001\/jamanetworkopen.2024.17641","article-title":"Performance of large language models on medical oncology examination questions","volume":"7","author":"Longwell","year":"2024","journal-title":"JAMA Netw Open"},{"key":"B7","doi-asserted-by":"publisher","DOI":"10.37047\/jos.galenos.2025.2024.106532","article-title":"Evaluation of 2024 turkish medical oncology board exam with chatgpt","volume":"11","author":"Kahraman","year":"2025","journal-title":"J Oncol Sci"},{"key":"B8","doi-asserted-by":"publisher","DOI":"10.1148\/radiol.230582","article-title":"Performance of ChatGPT on a radiology board-style examination: insights into current strengths and limitations","volume":"307","author":"Bhayana","year":"2023","journal-title":"Radiology"},{"key":"B9","doi-asserted-by":"publisher","DOI":"10.1016\/j.ajo.2023.05.024","article-title":"Performance of generative large language models on ophthalmology board-style questions","volume":"254","author":"Cai","year":"2023","journal-title":"Am J Ophthalmol"},{"key":"B10","doi-asserted-by":"publisher","first-page":"102938","DOI":"10.2139\/ssrn.4780937","article-title":"Medexpqa: multilingual benchmarking of large language models for medical question answering","volume":"155","author":"Alonso","year":"2024","journal-title":"Artif Intell Med"},{"key":"B11","doi-asserted-by":"publisher","DOI":"10.52202\/075280-2749","article-title":"C-eval: a multi-level multi-discipline Chinese evaluation suite for foundation models","volume":"36","author":"Huang","year":"2023","journal-title":"Adv Neural Inf Process Syst"},{"key":"B12","doi-asserted-by":"publisher","DOI":"10.3322\/caac.21834","article-title":"Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries","volume":"74","author":"Bray","year":"2024","journal-title":"CA: A Cancer J For Clin"},{"key":"B13","doi-asserted-by":"publisher","first-page":"64","DOI":"10.1016\/0030-4220(93)90436-8","article-title":"Nasopharyngeal carcinoma","volume":"394","author":"Chen","year":"2019","journal-title":"Lancet"},{"key":"B14","doi-asserted-by":"publisher","DOI":"10.1053\/j.gastro.2017.08.023","article-title":"Epidemiology of esophageal squamous cell carcinoma","volume":"154","author":"Abnet","year":"2018","journal-title":"Gastroenterology"},{"key":"B15","doi-asserted-by":"crossref","DOI":"10.18653\/v1\/2024.findings-acl.671","article-title":"Cmmlu: Measuring Massive Multitask Language Understanding in Chinese","author":"Li","year":"2024"},{"key":"B16","doi-asserted-by":"publisher","DOI":"10.1007\/s13187-025-02652-9","article-title":"A comparative analysis of GPT-4o and ERNIE Bot in a Chinese radiation oncology exam","volume":"41","author":"Wang","year":"2025","journal-title":"J Cancer Educ"},{"key":"B17","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2505.09388","article-title":"Qwen3 technical report","author":"Yang","year":"2025","journal-title":"arXiv"},{"key":"B18","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2512.02556","article-title":"Deepseek-v3.2: pushing the frontier of open large language models","author":"Liu","year":"2025","journal-title":"arXiv"},{"key":"B19","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2507.06261","article-title":"Gemini 2.5: pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities","author":"Comanici","year":"2025","journal-title":"arXiv"},{"key":"B20","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2601.03267","article-title":"OpenAI GPT-5 system card","author":"Singh","year":"2025","journal-title":"arXiv"},{"key":"B21","doi-asserted-by":"publisher","first-page":"141","DOI":"10.1038\/s41746-025-01533-1","article-title":"Medical foundation large language models for comprehensive text analysis and beyond","volume":"8","author":"Xie","year":"2025","journal-title":"NPJ Digital Med"},{"key":"B22","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2505.11462","article-title":"Disentangling reasoning and knowledge in medical large language models","author":"Thapa","year":"2025","journal-title":"arXiv"},{"key":"B23","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2212.09162","article-title":"Med-palm: large language models encode clinical knowledge","author":"Singhal","year":"2022","journal-title":"arXiv"},{"key":"B24","doi-asserted-by":"crossref","DOI":"10.18653\/v1\/D19-1259","article-title":"Pubmedqa: A Dataset for Biomedical Research Question Answering","author":"Jin","year":"2019"},{"key":"B25","doi-asserted-by":"publisher","DOI":"10.2196\/58375","article-title":"Performance of plug-in augmented ChatGPT and its ability to quantify uncertainty: simulation study on the German medical board examination","volume":"11","author":"Madrid","year":"2025","journal-title":"JMIR Med Educ"}],"container-title":["Frontiers in Oncology"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fonc.2026.1808714\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,6,17]],"date-time":"2026-06-17T04:52:00Z","timestamp":1781671920000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fonc.2026.1808714\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,6,17]]},"references-count":25,"alternative-id":["10.3389\/fonc.2026.1808714"],"URL":"https:\/\/doi.org\/10.3389\/fonc.2026.1808714","relation":{},"ISSN":["2234-943X"],"issn-type":[{"value":"2234-943X","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,6,17]]},"article-number":"1808714"}}