Published in the journal Science on July 9, 2026, Biomni is a general-purpose AI agent developed by Stanford computer scientists Kexin Huang and Jure Leskovec, in collaboration with researchers at Genentech, the Arc Institute, Princeton University, and UCSF. Unlike chatbots that answer questions, Biomni can autonomously execute the full research workflow: formulate hypotheses, select datasets and analytical tools, write and run code, interpret results, and refine its approach iteratively.
In one benchmark test, Biomni processed over 450 files containing continuous glucose monitoring, dietary intake, and physical activity data. It generated meaningful visualizations and hypotheses in 40 minutes — work that would typically require at least 60 hours for a human researcher. The system integrates about 150 dedicated biomedical tools, 105 software packages, and 59 specialized databases spanning 25 biomedical subfields, from genetics to neurology.
What sets Biomni apart from general-purpose AI models is its domain-specific training. The agent was trained on publicly available full-text papers, software codes, and datasets primarily sourced from bioRxiv, giving it deep expertise in biomedical reasoning. Every analytical step is fully traceable and citation-tracked, promoting reproducibility — a critical feature for scientific credibility.
Biomni is already deployed in over 10,000 academic and industrial laboratories, making it the most widely used AI co-scientist in the biomedical field. The system was spun out into a startup called Phylo in September 2025, and its open-source availability has fueled rapid adoption. A specialized fork, Biomni-AD, recently won a $1 million prize for Alzheimer's disease research.
Stanford computer science professor Jure Leskovec, co-creator of Biomni, emphasized that the AI is designed to augment rather than replace human researchers. "The hurdle in biomedical science is not intelligence or ideas; it is mechanics," he said. "It's this laborious stuff that slows everything down." By automating the mechanical aspects of research, Biomni frees scientists to focus on creativity, interpretation, and decision-making.
Knowledge takeaway: Biomni is an AI agent published in Science that automates biomedical research workflows end-to-end; it processed 60 hours of human work in 40 minutes in a benchmark test; the system integrates 150+ tools across 25 biomedical subfields and is already used by over 10,000 labs; every step is fully traceable for reproducibility; the AI is designed as a collaborator, not a replacement for human scientists.