What question did this study set out to answer?

The aim is to evaluate if large language models can assist scientists in synthesizing scientific literature.

February 7, 2026Open Access

Synthesizing scientific literature with retrieval-augmented language models

Key Points

The aim is to evaluate if large language models can assist scientists in synthesizing scientific literature.
Introduced OpenScholar, a retrieval-augmented language model to answer scientific queries.
Developed ScholarQABench, a benchmark for literature search with 2,967 queries and 208 answers across various domains.
Compared performance of OpenScholar-8B with GPT-4o and PaperQA2 in citation accuracy and correctness.
OpenScholar-8B outperformed GPT-4o by 6.1% and PaperQA2 by 5.5% in correctness on a synthesis task.
OpenScholar achieved citation accuracy comparable to human experts, in contrast to GPT-4o, which hallucinates citations 78–90% of the time.
Experts preferred OpenScholar responses over expert-written ones 51% and 70% of the time, respectively.

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) 1 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.

Read Full Paperexternally

Demander à l'IA

Bookmark

View Full Paper