New large quantitative models are changing structure‑based small‑molecule discovery: MIT and Recursion’s Boltz‑2 demonstrated rapid, accurate binding‑affinity predictions, while companies like SandboxAQ advocate physics‑grounded models to complement machine learning. Boltz‑2 topped CASP16 benchmarks and reports affinity estimates in seconds versus legacy free‑energy methods that take hours or days. Researchers and industry leaders say these advances could cut early‑stage screening costs and democratize computational hit‑finding, enabling more firms to move promising chemistries into synthesis and testing faster. The shift emphasizes hybrid approaches—physics plus data—that target affinity prediction rather than text‑trained models alone.