Lab-in-the-Loop at Compute Speed — Ian Quigley, Leash Bio
Leash Bio releases Minos, an open lab-in-the-loop benchmark enabling scientific RL for drug discovery agents.
“they discovered the data they had spent the last few years collecting could provide tens of thousands of "lab-in-the-loop" simulations, just what an agent needs for scientific RL”
Leash Bio has released Minos, a verifiable scientific reasoning environment built on their DNA Encoded Library (DEL) data, enabling reinforcement learning agents to run tens of thousands of lab-in-the-loop simulations. The release includes a preprint, blog post, and open dataset, positioning it as a potential benchmark for AI agents doing real chemistry. This matters because grounded, verifiable scientific environments for RL are a critical missing piece in AI-for-science research.