Researchers released PSBench, the largest curated collection of annotated protein structure models—1.4 million entries—with independent quality assessments to improve AI protein‑structure prediction workflows and benchmarking. In parallel, MIT chemical engineers reported an AI‑driven method to reprogram industrial yeast for protein production, cutting time and cost in biomanufacturing. PSBench aims to sharpen model evaluation for drug discovery applications; the MIT approach uses language‑model techniques to optimize strain engineering, promising faster, cheaper biologics production. Together these advances address both in silico design reliability and downstream manufacturing efficiency.