NVIDIA and the UW Medicine Institute for Protein Design released RFD4-Proteina, an open BioNeMo model that combines programmable protein design with search, letting scientists and agents generate targeted designs and refine them against molecular inputs, constraints and scoring objectives. In testing reported in a research draft, NVIDIA says the model reached higher computational design success rates and generation throughput than the evaluated baselines across four benchmark suites: protein binding, small-molecule binding, enzyme scaffolding and DNA binding. The company states that laboratory experiments remain what establish how a design performs under tested conditions. The model is available through partners including Amazon Bio Discovery, Anthropic's Claude Science, Benchling, Databricks, Google Cloud, Hugging Face, Nebius, OpenAI and Tamarind Bio. David Baker, the 2024 chemistry Nobel laureate who directs the institute, is quoted on moving from discovering proteins nature made to designing new ones.
Our takeCoupling generation to a search objective is the part that changes how the model gets used: it becomes something an agent can iterate against instead of a one-shot sampler. The performance claim rests on a research draft, with success rates reported against baselines the post does not name across four suites, so the ranking is only as good as that draft. Distribution through agent and cloud platforms is the more telling detail — that is where designed proteins leave the authors' own pipeline.