OpenScholar Outperforms ChatGPT in Scientific Citation Accuracy

University of Washington researchers develop specialized AI model for scientific literature search and synthesis.

Published on Feb. 9, 2026

Got story updates? Submit your updates here. ›

The landscape of artificial intelligence is rapidly evolving, with the development of highly specialized tools like OpenScholar, an open-source large language model (LLM) designed specifically for scientific literature. OpenScholar, created by researchers at the University of Washington, outperforms broader models like ChatGPT in citation accuracy and the usefulness of its research summaries. This signals a trend towards AI tailored to the nuances of specific disciplines, addressing concerns over the 'black box' nature and inaccuracies of general-purpose AI.

Why it matters

Researchers need tools they can trust to provide accurate summaries and identify relevant studies. OpenScholar's ability to deliver more comprehensive and detailed responses, rated as more useful over 50% of the time in manual evaluations by domain experts, highlights the value of this specialization. This isn't just about convenience; it's about accelerating the pace of scientific discovery.

The details

OpenScholar addresses the issues of general-purpose AI models by being trained exclusively on 45 million open access scientific papers. This focused training, combined with retrieval-augmented generation (RAG) – a technique that incorporates new information beyond its initial dataset – dramatically reduces errors and irrelevant citations. OpenScholar's open-source nature also allows for scrutiny, improvement, and customization by the wider scientific community, fostering trust and accelerating innovation.

  • OpenScholar was developed in 2026 by researchers at the University of Washington.

The players

OpenScholar

An open-source large language model (LLM) designed specifically for scientific literature search and synthesis.

Hannaneh Hajishirzi

A computer scientist at the University of Washington who co-developed OpenScholar.

Akari Asai

A computer scientist at the University of Washington who co-developed OpenScholar.

ChatGPT

A general-purpose language model developed by OpenAI that has captured public attention, but has been criticized for its 'black box' nature and tendency towards inaccuracies.

Got photos? Submit your photos here. ›

What they’re saying

“The focus is now on ensuring the correctness of answers – a key concern with general-purpose AI.”

— Hannaneh Hajishirzi, Co-developer of OpenScholar

What’s next

The University of Washington team is already pushing the boundaries of AI-powered research with the development of Deep Research Tulu, aiming for even more comprehensive scientific responses.

The takeaway

OpenScholar's open-source and specialized approach to scientific literature search and synthesis highlights the growing trend towards AI models tailored to specific disciplines, addressing the limitations of general-purpose AI and accelerating the pace of scientific discovery.