Google’s DeepMind announced that AlphaFold 2.5 reached 95 percent accuracy in blind protein-folding tests conducted with the European Bioinformatics Institute. The update uses reinforcement learning to model large protein complexes and rare folds faster than earlier versions. The scientific community hailed the breakthrough as a major step toward AI-driven drug design. 

Biotech investments in computational biology spiked 4 percent this week following the announcement. DeepMind plans to open-source model weights for academic use in early 2026 under its AlphaFold Database initiative.