Gabriele Corso

Gabriele Corso is turning biomolecular AI from a guarded capability into shared scientific infrastructure—without mistaking benchmark promise for a finished drug. His work combines model research, permissive release and institution-building around a field whose claims demand unusually careful validation.

Drug discovery is often described as a race for better models. Gabriele Corso has made it a contest over who gets to use them. An Italian computer scientist trained at Cambridge and MIT, he helped lead the Boltz family of open biomolecular models and now runs a public-benefit company built around open research. His defining 2025 contribution, Boltz-2, brings together two problems often treated apart: predicting how biomolecules arrange themselves and estimating how strongly they bind. The project reported performance approaching free-energy perturbation on selected affinity benchmarks, with substantially lower computational requirements, and released model weights, inference code and training code under a permissive licence. That made an advanced molecular-modelling system inspectable and adaptable beyond a small circle of well-capitalised laboratories. The openness matters as much as the benchmark. Drug discovery is cumulative: models improve when scientists can test difficult targets, expose failure modes and build upon one another’s work. Corso treats access not as a communications gesture but as part of the scientific method. The ambition is considerable, but so is the need for restraint. Boltz-2’s comparative accuracy remains under active scrutiny, and computational promise is not a therapy. What is already clear is Corso’s unusual ability to move between mathematical research, usable scientific infrastructure and institution-building while keeping the frontier open to examination.