Researchers at the University of Washington and the Allen Institute for AI developed OpenScholar, an open‑source AI model designed to accurately synthesize and cite current scientific research. Unlike general‑purpose AI systems that often hallucinate sources, OpenScholar grounds its answers in a database of 45 million research papers and uses retrieval‑augmented generation to incorporate new findings.
In tests published in Nature, OpenScholar cited sources as accurately as human experts, and scientists preferred its responses over expert‑written answers 51% of the time. The team also created a new benchmark, ScholarQABench, to evaluate research synthesis, and has made OpenScholar’s code, data and demo freely available to support transparent AI for science.