Scientists unveil AI that creates never-before-seen molecules to spee…

Researchers at Rovira i Virgili University in Spain have developed CoCoGraph, an AI system using diffusion models to generate chemically valid molecules at scale, achieving 40% human expert deception rates and outperforming competing tools on two-thirds of physicochemical property tests. The breakthrough could accelerate drug discovery and materials science by systematically exploring the vast chemical space of 10^60 possible molecules. If you're building drug discovery software or screening tools, your moat just shrank: generative chemistry is now good enough that humans can't tell AI molecules from real ones 40% of the time, and it runs on less compute than alternatives. Your next customer won't need you to find molecules—they'll need you to validate, synthesise, or optimise what an AI generates.