{"database":"biostudies-literature","file_versions":[],"scores":null,"additional":{"omics_type":["Unknown"],"volume":["650(8103)"],"submitter":["Asai A"],"pubmed_abstract":["Scientific progress depends on the ability of researchers to synthesize the growing body of literature. Can large language models (LLMs) assist scientists in this task? Here we introduce OpenScholar, a specialized retrieval-augmented language model (LM)<sup>1</sup> that answers scientific queries by identifying relevant passages from 45 million open-access papers and synthesizing citation-backed responses. To evaluate OpenScholar, we develop ScholarQABench, the first large-scale multi-domain benchmark for literature search, comprising 2,967 expert-written queries and 208 long-form answers across computer science, physics, neuroscience and biomedicine. Despite being a smaller open model, OpenScholar-8B outperforms GPT-4o by 6.1% and PaperQA2 by 5.5% in correctness on a challenging multi-paper synthesis task from the new ScholarQABench. Although GPT-4o hallucinates citations 78-90% of the time, OpenScholar achieves citation accuracy on par with human experts. OpenScholar's data store, retriever and self-feedback inference loop improve off-the-shelf LMs: for instance, OpenScholar-GPT-4o improves the correctness of GPT-4o by 12%. In human evaluations, experts preferred OpenScholar-8B and OpenScholar-GPT-4o responses over expert-written ones 51% and 70% of the time, respectively, compared with 32% for GPT-4o. We open-source all artefacts, including our code, models, data store, datasets and a public demo."],"journal":["Nature"],"pagination":["857-863"],"full_dataset_link":["https://www.ebi.ac.uk/biostudies/studies/S-EPMC12935541"],"repository":["biostudies-literature"],"pubmed_title":["Synthesizing scientific literature with retrieval-augmented language models."],"pmcid":["PMC12935541"],"pubmed_authors":["Koh PW","Neubig G","Shi W","Wu B","Liu S","Xiong Y","Latzke M","Yih WT","D'Arcy M","Feldman S","Hajishirzi H","Tong H","Shao R","Zettlemoyer L","Tian M","He J","Soldaini L","Downey D","Singh A","Lo K","Ji P","Wadden D","Hwang JD","Asai A","Sparks J","Kishore V","Chang JC","Weld DS"],"additional_accession":[]},"is_claimable":false,"name":"Synthesizing scientific literature with retrieval-augmented language models.","description":"Scientific progress depends on the ability of researchers to synthesize the growing body of literature. Can large language models (LLMs) assist scientists in this task? Here we introduce OpenScholar, a specialized retrieval-augmented language model (LM)<sup>1</sup> that answers scientific queries by identifying relevant passages from 45 million open-access papers and synthesizing citation-backed responses. To evaluate OpenScholar, we develop ScholarQABench, the first large-scale multi-domain benchmark for literature search, comprising 2,967 expert-written queries and 208 long-form answers across computer science, physics, neuroscience and biomedicine. Despite being a smaller open model, OpenScholar-8B outperforms GPT-4o by 6.1% and PaperQA2 by 5.5% in correctness on a challenging multi-paper synthesis task from the new ScholarQABench. Although GPT-4o hallucinates citations 78-90% of the time, OpenScholar achieves citation accuracy on par with human experts. OpenScholar's data store, retriever and self-feedback inference loop improve off-the-shelf LMs: for instance, OpenScholar-GPT-4o improves the correctness of GPT-4o by 12%. In human evaluations, experts preferred OpenScholar-8B and OpenScholar-GPT-4o responses over expert-written ones 51% and 70% of the time, respectively, compared with 32% for GPT-4o. We open-source all artefacts, including our code, models, data store, datasets and a public demo.","dates":{"release":"2026-01-01T00:00:00Z","publication":"2026 Feb","modification":"2026-07-11T03:18:15.057Z","creation":"2026-07-11T03:12:04.547Z"},"accession":"S-EPMC12935541","cross_references":{"pubmed":["41639446"],"doi":["10.1038/s41586-025-10072-4"]}}