Biomni: General Biomedical AI Agent Published in Science

On July 10, the general biomedical AI agent Biomni was officially published in the journal Science. The system marks that life science research is rapidly moving towards a new stage autonomously driven by AI, attracting widespread attention from academia and industry.

Core Capabilities and Technical Details

Biomni is a general AI agent integrated with a biology environment, capable of reasoning, planning, and executing end-to-end research workflows. In various evaluation tasks, its performance surpassed comparative Large Language Models (LLMs). Biomni's capabilities cover a wide range of biomedical fields, including drug repurposing, rare disease diagnosis, single-cell annotation, causal gene identification, molecular cloning, protein stability design, and coordinating wet lab experiments. Additionally, it can automate large-scale omics analysis, coordinate laboratory robots, optimize molecular properties, and help train biology-specific models.

Development Trajectory and Industry Impact

According to relevant discussions, Biomni has a clear timeline: the research appeared on arXiv 13 months ago and was open-sourced on GitHub 11 months ago. Scholars like Eric Topol believe that such systems can collaborate deeply with human researchers in the future, accelerating biological discoveries from basic research to translational applications. Author Kexin Huang pointed out that since access to frontier models is difficult in biomedical scenarios and frontier labs are closer to direct drug development, biopharmaceutical companies increasingly need "bio-native" agent capabilities that are not tied to a single model provider and can fully leverage open-source models. This indicates that life science agents have practically moved towards commercialization and corporate application.

2026-07-10 ~ 2026-07-10 · 10 related posts

3 near-duplicate retellings: KexinHuang5 · ChengleiSi · KexinHuang5