EMERYVILLE, Calif.— Profluent, an AI-driven protein design, has announced the release of ProGen3, a groundbreaking suite of protein-generating foundation models.
Trained on over 3.4 billion protein sequences, ProGen3 is the first to demonstrate wet lab evidence of the scaling benefits in biology akin to those seen in large language models.
The models, ranging from 112 million to 46 billion parameters, are built on a 1.5-trillion-token dataset and validated through lab experiments. Results show that as the models scale, their ability to design functional proteins improves significantly. This includes single-shot generation of antibodies for 20 major drug targets—therapies that have reached 7 million patients and represent $660 billion in historical sales.
ProGen3 also powers the design of ultra-compact gene editors like OpenCRISPR-1, the world’s first AI-created and open-source gene editor. These innovations are available for licensing through Profluent’s platform, with exclusive early access offered to select partners across therapeutics, agriculture, and biomanufacturing.
“This is just the starting line for AI’s impact on biology,” said CEO Ali Madani, Ph.D. “With ProGen3, we’re moving from chance discovery to intentional design.”
Profluent’s platform aims to reshape biotech R&D by making biology as programmable as language.
