Overview
- Researchers outlined Biomni in Science as an LLM-powered “co‑scientist” built on a curated action space that includes 150 specialized tools, 105 software packages, and 59 databases.
- Biomni’s agent (Biomni‑A1) retrieves relevant tools and data, reasons with a large language model, and outputs each step as executable code so workflows can be run or adapted automatically.
- The team validated the system across five case studies — from wearable sensor analysis and large‑scale single‑cell genomics to lab protocol design, protein optimization, and robotic orchestration — and found performance comparable to experts.
- In one example Biomni analyzed more than 450 real‑world wearable files and generated plausible hypotheses in about 40 minutes versus an estimated 60+ human hours, illustrating the reported speed gains.
- The authors say a prototype is being used in over 10,000 academic and industry labs, but independent, broader validation and discussion of governance, reproducibility, and safety are not detailed in the current reports.