The Baker lab at the University of Washington released RFdiffusion3 as open source, unveiling an all-atom de novo protein design model that can generate binders and catalytic enzymes that interact with DNA, small molecules and proteins. The preprint authored by Rohith Krishna and colleagues (not yet peer reviewed) reports experimental proof-of-concept for designing DNA-recognizing proteins and complex enzymatic architectures. RFdiffusion3 advances the prior RFdiffusion2 approach by designing at the atomic level and delivering roughly ten-fold faster performance, the team says. The model’s generality supports potential applications spanning synthetic transcription factors for gene therapy to industrial biocatalysis. Open release will let academic and industry groups iterate on design workflows; it also intensifies competition among AI-driven protein-design platforms and could speed translation of novel modalities into preclinical pipelines.