Profluent, an AI-driven protein design company, has introduced ProGen3, a new family of advanced protein models trained on the world’s largest curated protein sequence dataset. The company says its models demonstrate that scaling principles—widely observed in natural language processing—also apply to biological design. Profluent’s ProGen3 models range up to 46 billion parameters and have been trained on 1.5 trillion tokens, enabling high-performance protein generation. The company validated its findings through laboratory testing, showing that larger models produce more effective proteins and better align with experimental data. The ProGen3 family builds on Profluent’s growing database of 80 billion protein sequences, significantly exceeding existing public resources.
Profluent has already applied earlier versions of its models to notable achievements, including OpenCRISPR-1, an AI-designed and open-source gene editor. Its platform has also produced antibody designs, called OpenAntibodies, which reportedly match or outperform existing therapies across 20 disease targets. Profluent offers its technology through licensing, custom model access, and strategic partnerships across medicine, agriculture, and manufacturing. CEO Dr. Ali Madani described, “We view this as the starting line in the race toward emergent capabilities in biological design and look forward to working with our partners to actualize the potential of our protein design technology.” The company, based in Emeryville, California, is supported by investors including Spark Capital and Insight Partners.
