An international team from the University of Washington, the University of Oxford and Nvidia published the full design of RFdiffusion3, a generative AI system that designs proteins and nucleic acids at the single-atom level, in Science on the 16th local time; the team had open-sourced the system on December 3, 2025. RFdiffusion3 draws on the AlphaFold3 network architecture, represents each amino acid with up to 14 atoms and designs structures by directly adjusting atomic 3D coordinates, generating proteins that satisfy constraints such as specified reactive-atom positions, hydrogen-bond conditions and allowed atomic distributions. Per IT之家, RFdiffusion3 outperformed RFdiffusion2 at 37 of 41 enzyme active-site comparisons and generated structures about ten times faster, while its designed paraoxonase and peptide-cleaving enzymes completed the intended reactions in real experiments. The study's corresponding author is University of Washington professor David Baker, a 2024 Nobel chemistry laureate.