Researchers at MIT, with Recursion, released Boltz‑2 — an open‑source model that markedly improved binding‑affinity prediction speed and accuracy and topped CASP16 affinity benchmarks. Boltz‑2 reportedly computes binding affinities in about 20 seconds, a dramatic acceleration versus free‑energy perturbation simulations, potentially allowing teams to triage and optimize small‑molecule candidates far faster. Industry voices, including SandboxAQ, emphasize that physics‑grounded large quantitative models (LQMs) and structural co‑folding advances from DeepMind and Baker lab work are expanding computational chemistry’s role in discovery. While the models compress early‑stage screening timelines, firms still pair them with experimental validation and proprietary datasets to drive candidate selection.