{"url":"/sota/question-answering-on-medqa-usmle","task":{"name":"Question Answering","url":"/task/question-answering","note":null},"dataset":{"name":"MedQA","url":"/dataset/medqa-usmle"},"category":"Natural Language Processing","categories":["Miscellaneous","Natural Language Processing","Reasoning"],"category_note":null,"description":"Question answering can be segmented into domain-specific tasks like community question answering and knowledge-base question answering. Popular benchmark datasets for evaluation question answering systems include [SQuAD](/dataset/squad), [HotPotQA](/dataset/hotpotqa), [bAbI](/dataset/babi-1), [TriviaQA](/dataset/triviaqa), [WikiQA](/dataset/wikiqa), and many others. Models for question answering are typically evaluated on metrics like EM and F1. Some recent top performing models are T5 and XLNet.\r\n\r\n( Image credit: [SQuAD](https://rajpurkar.github.io/mlx/qa-and-squad/) )","description_from":"task","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","rank":"the archive's row order at snapshot; not re-ranked","rows_end_at":"2025-07-28","rows_withheld_as_spam":0,"metric_values":"the archive's strings, untouched"},"metrics":["Accuracy"],"metric_direction":{"note":"inferred from the metric name only (the archive records no direction); null = not inferred, chart draws points only","by_metric":{"Accuracy":"higher"}},"counts":{"rows":27,"rows_with_code":22,"rows_with_paper_page":27,"rows_dated":27,"rows_using_additional_data":7},"rows":[{"rank_in_archive_order":1,"model":"Med-Gemini","metrics":{"Accuracy":"91.1"},"uses_additional_data":true,"paper_date":"2024-04-29","paper":"/paper/capabilities-of-gemini-models-in-medicine","paper_url":"https://arxiv.org/abs/2404.18416v2","paper_title":"Capabilities of Gemini Models in Medicine","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":2,"model":"GPT-4","metrics":{"Accuracy":"90.2"},"uses_additional_data":true,"paper_date":"2023-11-28","paper":"/paper/can-generalist-foundation-models-outcompete","paper_url":"https://arxiv.org/abs/2311.16452v1","paper_title":"Can Generalist Foundation Models Outcompete Special-Purpose Tuning? Case Study in Medicine","code":"https://github.com/microsoft/promptbase","n_code_links":2,"syntology":{"n_ran":9,"n_unverified":1,"n_samples":10,"n_pointer_only_licence":0}},{"rank_in_archive_order":3,"model":"Med-PaLM 2","metrics":{"Accuracy":"85.4"},"uses_additional_data":false,"paper_date":"2023-05-16","paper":"/paper/towards-expert-level-medical-question","paper_url":"https://arxiv.org/abs/2305.09617v1","paper_title":"Towards Expert-Level Medical Question Answering with Large Language Models","code":"https://github.com/m42-health/med42","n_code_links":1,"syntology":null},{"rank_in_archive_order":4,"model":"Med-PaLM 2 (CoT + SC)","metrics":{"Accuracy":"83.7"},"uses_additional_data":false,"paper_date":"2023-05-16","paper":"/paper/towards-expert-level-medical-question","paper_url":"https://arxiv.org/abs/2305.09617v1","paper_title":"Towards Expert-Level Medical Question Answering with Large Language Models","code":"https://github.com/m42-health/med42","n_code_links":1,"syntology":null},{"rank_in_archive_order":5,"model":"Med-PaLM 2 (5-shot)","metrics":{"Accuracy":"79.7"},"uses_additional_data":false,"paper_date":"2023-05-16","paper":"/paper/towards-expert-level-medical-question","paper_url":"https://arxiv.org/abs/2305.09617v1","paper_title":"Towards Expert-Level Medical Question Answering with Large Language Models","code":"https://github.com/m42-health/med42","n_code_links":1,"syntology":null},{"rank_in_archive_order":6,"model":"MedMobile (3.8B)","metrics":{"Accuracy":"75.7"},"uses_additional_data":true,"paper_date":"2024-10-11","paper":"/paper/medmobile-a-mobile-sized-language-model-with","paper_url":"https://arxiv.org/abs/2410.09019v1","paper_title":"MedMobile: A mobile-sized language model with expert-level clinical capabilities","code":"https://github.com/nyuolab/MedMobile","n_code_links":1,"syntology":null},{"rank_in_archive_order":7,"model":"Meerkat-7B","metrics":{"Accuracy":"74.3"},"uses_additional_data":true,"paper_date":"2024-03-30","paper":"/paper/small-language-models-learn-enhanced","paper_url":"https://arxiv.org/abs/2404.00376v2","paper_title":"Small Language Models Learn Enhanced Reasoning Skills from Medical Textbooks","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":8,"model":"Meerkat-7B (Single)","metrics":{"Accuracy":"70.6"},"uses_additional_data":true,"paper_date":"2024-03-30","paper":"/paper/small-language-models-learn-enhanced","paper_url":"https://arxiv.org/abs/2404.00376v2","paper_title":"Small Language Models Learn Enhanced Reasoning Skills from Medical Textbooks","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":9,"model":"Meditron-70B (CoT + SC)","metrics":{"Accuracy":"70.2"},"uses_additional_data":false,"paper_date":"2023-11-27","paper":"/paper/meditron-70b-scaling-medical-pretraining-for","paper_url":"https://arxiv.org/abs/2311.16079v1","paper_title":"MEDITRON-70B: Scaling Medical Pretraining for Large Language Models","code":"https://github.com/epfllm/meditron","n_code_links":1,"syntology":{"n_ran":9,"n_unverified":5,"n_samples":14,"n_pointer_only_licence":0}},{"rank_in_archive_order":10,"model":"Flan-PaLM (540 B)","metrics":{"Accuracy":"67.6"},"uses_additional_data":false,"paper_date":"2022-12-26","paper":"/paper/large-language-models-encode-clinical","paper_url":"https://arxiv.org/abs/2212.13138v1","paper_title":"Large Language Models Encode Clinical Knowledge","code":"https://github.com/dmis-lab/olaph","n_code_links":1,"syntology":null},{"rank_in_archive_order":11,"model":"LLAMA-2 (70B SC CoT)","metrics":{"Accuracy":"61.5"},"uses_additional_data":true,"paper_date":"2023-11-27","paper":"/paper/meditron-70b-scaling-medical-pretraining-for","paper_url":"https://arxiv.org/abs/2311.16079v1","paper_title":"MEDITRON-70B: Scaling Medical Pretraining for Large Language Models","code":"https://github.com/epfllm/meditron","n_code_links":1,"syntology":{"n_ran":9,"n_unverified":5,"n_samples":14,"n_pointer_only_licence":0}},{"rank_in_archive_order":12,"model":"Shakti-LLM (2.5B)","metrics":{"Accuracy":"60.3"},"uses_additional_data":false,"paper_date":"2024-10-15","paper":"/paper/shakti-a-2-5-billion-parameter-small-language","paper_url":"https://arxiv.org/abs/2410.11331v1","paper_title":"SHAKTI: A 2.5 Billion Parameter Small Language Model Optimized for Edge AI and Low-Resource Environments","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":13,"model":"Codex 5-shot CoT","metrics":{"Accuracy":"60.2"},"uses_additional_data":false,"paper_date":"2022-07-17","paper":"/paper/can-large-language-models-reason-about","paper_url":"https://arxiv.org/abs/2207.08143v4","paper_title":"Can large language models reason about medical questions?","code":"https://github.com/vlievin/medical-reasoning","n_code_links":1,"syntology":{"n_ran":10,"n_unverified":0,"n_samples":10,"n_pointer_only_licence":0}},{"rank_in_archive_order":14,"model":"LLAMA-2 (70B)","metrics":{"Accuracy":"59.2"},"uses_additional_data":true,"paper_date":"2023-11-27","paper":"/paper/meditron-70b-scaling-medical-pretraining-for","paper_url":"https://arxiv.org/abs/2311.16079v1","paper_title":"MEDITRON-70B: Scaling Medical Pretraining for Large Language Models","code":"https://github.com/epfllm/meditron","n_code_links":1,"syntology":{"n_ran":9,"n_unverified":5,"n_samples":14,"n_pointer_only_licence":0}},{"rank_in_archive_order":15,"model":"VOD (BioLinkBERT)","metrics":{"Accuracy":"55.0"},"uses_additional_data":false,"paper_date":"2022-09-23","paper":"/paper/variational-open-domain-question-answering","paper_url":"https://arxiv.org/abs/2210.06345v2","paper_title":"Variational Open-Domain Question Answering","code":"https://github.com/VodLM/vod","n_code_links":2,"syntology":{"n_ran":1,"n_unverified":0,"n_samples":1,"n_pointer_only_licence":0}},{"rank_in_archive_order":16,"model":"BioMedGPT-10B","metrics":{"Accuracy":"50.4"},"uses_additional_data":false,"paper_date":"2023-08-18","paper":"/paper/biomedgpt-open-multimodal-generative-pre","paper_url":"https://arxiv.org/abs/2308.09442v2","paper_title":"BioMedGPT: Open Multimodal Generative Pre-trained Transformer for BioMedicine","code":"https://github.com/pharmolix/openbiomed","n_code_links":1,"syntology":null},{"rank_in_archive_order":17,"model":"PubMedGPT (2.7 B)","metrics":{"Accuracy":"50.3"},"uses_additional_data":false,"paper_date":"2022-12-26","paper":"/paper/large-language-models-encode-clinical","paper_url":"https://arxiv.org/abs/2212.13138v1","paper_title":"Large Language Models Encode Clinical Knowledge","code":"https://github.com/dmis-lab/olaph","n_code_links":1,"syntology":null},{"rank_in_archive_order":18,"model":"DRAGON + BioLinkBERT","metrics":{"Accuracy":"47.5"},"uses_additional_data":false,"paper_date":"2022-10-17","paper":"/paper/deep-bidirectional-language-knowledge-graph","paper_url":"https://arxiv.org/abs/2210.09338v2","paper_title":"Deep Bidirectional Language-Knowledge Graph Pretraining","code":"https://github.com/michiyasunaga/dragon","n_code_links":2,"syntology":{"n_ran":3,"n_unverified":15,"n_samples":18,"n_pointer_only_licence":0}},{"rank_in_archive_order":19,"model":"BioLinkBERT (340 M)","metrics":{"Accuracy":"45.1"},"uses_additional_data":false,"paper_date":"2022-12-26","paper":"/paper/large-language-models-encode-clinical","paper_url":"https://arxiv.org/abs/2212.13138v1","paper_title":"Large Language Models Encode Clinical Knowledge","code":"https://github.com/dmis-lab/olaph","n_code_links":1,"syntology":null},{"rank_in_archive_order":20,"model":"GAL 120B (zero-shot)","metrics":{"Accuracy":"44.4"},"uses_additional_data":false,"paper_date":"2022-11-16","paper":"/paper/galactica-a-large-language-model-for-science-1","paper_url":"https://arxiv.org/abs/2211.09085v1","paper_title":"Galactica: A Large Language Model for Science","code":"https://github.com/paperswithcode/galai","n_code_links":1,"syntology":{"n_ran":2,"n_unverified":0,"n_samples":2,"n_pointer_only_licence":0}},{"rank_in_archive_order":21,"model":"BioLinkBERT (base)","metrics":{"Accuracy":"40.0"},"uses_additional_data":false,"paper_date":"2022-03-29","paper":"/paper/linkbert-pretraining-language-models-with","paper_url":"https://arxiv.org/abs/2203.15827v1","paper_title":"LinkBERT: Pretraining Language Models with Document Links","code":"https://github.com/michiyasunaga/LinkBERT","n_code_links":1,"syntology":{"n_ran":3,"n_unverified":11,"n_samples":14,"n_pointer_only_licence":0}},{"rank_in_archive_order":22,"model":"GrapeQA: PEGA","metrics":{"Accuracy":"39.51"},"uses_additional_data":false,"paper_date":"2023-03-22","paper":"/paper/grapeqa-graph-augmentation-and-pruning-to","paper_url":"https://arxiv.org/abs/2303.12320v2","paper_title":"GrapeQA: GRaph Augmentation and Pruning to Enhance Question-Answering","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":23,"model":"BioBERT (large)","metrics":{"Accuracy":"36.7"},"uses_additional_data":false,"paper_date":"2019-01-25","paper":"/paper/biobert-a-pre-trained-biomedical-language","paper_url":"https://arxiv.org/abs/1901.08746v4","paper_title":"BioBERT: a pre-trained biomedical language representation model for biomedical text mining","code":"https://github.com/dmis-lab/biobert","n_code_links":19,"syntology":{"n_ran":13,"n_unverified":12,"n_samples":25,"n_pointer_only_licence":1}},{"rank_in_archive_order":24,"model":"BioBERT (base)","metrics":{"Accuracy":"34.1"},"uses_additional_data":false,"paper_date":"2019-01-25","paper":"/paper/biobert-a-pre-trained-biomedical-language","paper_url":"https://arxiv.org/abs/1901.08746v4","paper_title":"BioBERT: a pre-trained biomedical language representation model for biomedical text mining","code":"https://github.com/dmis-lab/biobert","n_code_links":19,"syntology":{"n_ran":13,"n_unverified":12,"n_samples":25,"n_pointer_only_licence":1}},{"rank_in_archive_order":25,"model":"GPT-Neo (2.7 B)","metrics":{"Accuracy":"33.3"},"uses_additional_data":false,"paper_date":"2022-12-26","paper":"/paper/large-language-models-encode-clinical","paper_url":"https://arxiv.org/abs/2212.13138v1","paper_title":"Large Language Models Encode Clinical Knowledge","code":"https://github.com/dmis-lab/olaph","n_code_links":1,"syntology":null},{"rank_in_archive_order":26,"model":"BLOOM (few-shot, k=5)","metrics":{"Accuracy":"23.3"},"uses_additional_data":false,"paper_date":"2022-11-16","paper":"/paper/galactica-a-large-language-model-for-science-1","paper_url":"https://arxiv.org/abs/2211.09085v1","paper_title":"Galactica: A Large Language Model for Science","code":"https://github.com/paperswithcode/galai","n_code_links":1,"syntology":{"n_ran":2,"n_unverified":0,"n_samples":2,"n_pointer_only_licence":0}},{"rank_in_archive_order":27,"model":"OPT (few-shot, k=5)","metrics":{"Accuracy":"22.8"},"uses_additional_data":false,"paper_date":"2022-11-16","paper":"/paper/galactica-a-large-language-model-for-science-1","paper_url":"https://arxiv.org/abs/2211.09085v1","paper_title":"Galactica: A Large Language Model for Science","code":"https://github.com/paperswithcode/galai","n_code_links":1,"syntology":{"n_ran":2,"n_unverified":0,"n_samples":2,"n_pointer_only_licence":0}}],"since_archive":{"claim":"Results that newer papers report for their own method, placed here by Syntology. A model pointed at the cell in the paper's own table; the number was read from that cell and checked against this leaderboard's metric, dataset, split and scale; an independent check that saw this leaderboard's other rows and every other leaderboard on the same dataset accepted it. Not reviewed by the paper's authors or by the archive's editors, and not ranked against the archive rows.","extraction_file_present":true,"measurement":{"test_papers":883,"papers_with_output":881,"judged_true":108,"judged":110,"wilson95_lower":0.9361,"measured_on":"2026-09-24","frozen_commit":"0e3de0df94"},"measurement_note":"blind adjudication of accepted entries on a held-out split of archive papers, rules frozen before the test","coverage":{"sentence":"Syntology has checked 6,885 of the 9,623 papers on this site that are newer than the archive; results from the others appear after they are checked.","complete":false,"papers_newer_than_archive":9623,"papers_checked":6885,"papers_extracted_not_yet_verified":0,"boards_without_verdict":2,"papers_not_yet_extracted":2737},"order":"newest first by month (arXiv date, else the arXiv-id month), then arXiv id descending","columns":["Accuracy"],"entries":[{"paper":"/paper/arxiv-2607-17266","paper_title":"Debate-on-Graph: Reliable and Adaptive Reasoning of Large Language Model on Uncertain Knowledge Graph","arxiv_id":"2607.17266","model":"DoG (Ours) · GPT","row_label":"DoG (Ours)","configuration":"GPT","configuration_in_model":true,"configuration_not_shown_because":null,"values":{"Accuracy":"81.5"},"date":"2026-07-19","month_from_arxiv_id":null,"where_in_paper":"Table 2, row “DoG (Ours)”","table_label":"2","table_label_from":"the caption's own label","table_position_1based":2,"caption_head":"Table 2 : Accuracy (%) comparison of DoG and DoG-E with different baselines on…","cells":[{"column":"Accuracy","cell_text":"81.5","column_header":"MedQA / GPT","value":81.5,"table_index_0based":1,"row":13,"col":8}],"code":"3 of 3 ran"},{"paper":"/paper/arxiv-2605-23414","paper_title":"When Planning Fails Despite Correct Execution: On Epistemic Calibration for LLM-Based Multi-Agent Systems","arxiv_id":"2605.23414","model":"EPC-AW","row_label":"EPC-AW","configuration":"Acc.","configuration_in_model":false,"configuration_not_shown_because":"names_the_metric","values":{"Accuracy":"74.11"},"date":"2026-05-22","month_from_arxiv_id":null,"where_in_paper":"Table 1, row “EPC-AW”","table_label":"1","table_label_from":"the caption's own label","table_position_1based":1,"caption_head":"Table 1 : Accuracy (%) comparison of different repair methods. \\Delta denotes…","cells":[{"column":"Accuracy","cell_text":"74.11","column_header":"MedQA / Acc.","value":74.11,"table_index_0based":0,"row":5,"col":11}],"code":"2 of 2 ran"}]},"syntology":{"read_at":"2026-09-25T09:33:49+00:00","claim":"Per row: N of M harvested code samples from that row's paper executed on a synthesized fixture; the other M-N are unverified. Not a reproduction of the row's number; not a correctness claim. n_pointer_only_licence counts samples the site points at rather than redistributes (a licence axis, independent of ran/unverified).","rows_with_graph_line":13,"rows_with_any_sample_ran":13,"distinct_papers_with_graph_line":8,"distinct_papers_with_any_sample_ran":8,"samples_over_distinct_papers":{"n_ran":50,"n_unverified":44,"n_samples":94,"n_pointer_only_licence":1,"note":"each paper (arXiv id) counted once, however many rows it is behind; this is the page-level figure"},"samples_row_weighted":{"n_ran":85,"n_unverified":66,"n_samples":151,"n_pointer_only_licence":2,"note":"row-weighted: a paper behind several rows is counted once per row; inflated relative to samples_over_distinct_papers by design, kept for readers summing the per-row syntology blocks"}}}