Academic groups and startups continue to push AI into core drug‑discovery tasks: MIT and Recursion released Boltz‑2 for rapid binding‑affinity prediction, and SandboxAQ and others are advancing physics‑based large quantitative models to simulate molecular systems. These tools aim to reduce reliance on costly experimental screens and shorten lead‑optimization timelines. In parallel, reports claim Google’s new foundation models proposed and validated a novel cancer drug combination, highlighting big‑tech capabilities to generate actionable hypotheses. Both open‑source and proprietary models are now demonstrating predictive performance in structure and binding tasks, prompting commercial teams to integrate models into lead selection and compound design. The field still requires careful benchmarking and prospective validation, but the combination of open models, physics grounding, and infrastructure scale is changing early discovery economics and lowering barriers for small teams to run high‑value in silico campaigns.