KAIST Unveils K-Fold, a Bio AI Model That Outperforms Google’s AlphaFold3 — BigGo Finance

KAIST has unveiled K-Fold, a next-generation bio AI model that accelerates protein structure prediction by up to 25 times compared to existing models. The university announced that K-Fold outperformed Google DeepMind’s AlphaFold3 in certain evaluation metrics.

KAIST announced on the 28th that “Team KAIST,” the lead institution in South Korea’s Ministry of Science and ICT’s “AI-Specialized Foundation Model Development Program,” developed K-Fold. Professor Kim Woo-youn of the Department of Chemistry led the research, while Professor Hwang Sung-joo and Professor Ahn Sung-soo of the Kim Jaechul Graduate School of AI developed the AI model, and Professors Oh Byung-ha, Kim Ho-min, and Lee Kyu-ri of the Department of Biological Sciences handled protein data construction and validation.

The core of K-Fold lies in predicting protein-drug “binding,” an essential step in drug development. Beyond predicting three-dimensional protein structures, it calculates where and how drug candidates will bind to proteins, enabling rapid screening of promising drug candidates. Its prediction scope extends beyond protein-protein and protein-drug candidate interactions to include complex structures formed when multiple biomolecules such as DNA and RNA bind together.

The performance evaluation results are noteworthy. In a March phase assessment, K-Fold’s molecular complex structure prediction accuracy was rated as approaching AlphaFold3. In an internal performance evaluation conducted by the research team in August, K-Fold demonstrated higher performance than existing global models in certain metrics. It showed particularly strong predictive performance in G-protein coupled receptors (GPCRs)—major drug targets for cancer and other diseases—as well as kinases and targeted protein degradation (TPD), which directly eliminates disease-causing proteins.

The secret to its speed improvement lies in eliminating pre-computation steps. Existing protein structure prediction AI models required complex preliminary work involving large-scale searching and comparison of similar protein sequence information before structural calculations. K-Fold applies a new approach that does not rely on this process, increasing structure prediction speed by up to 25 times compared to existing models. This allows researchers to evaluate more drug candidates in the same timeframe, helping narrow down targets for actual experimentation.

K-Fold has also been implemented as an AI service for practical research use. It is integrated into HyperLab, a multi-agent platform developed by HITS, a KAIST faculty startup. Researchers can conduct drug discovery through conversational AI interactions on the web without building dedicated high-performance computing infrastructure or managing complex AI programs.

HyperLab features over 120 computational tools and more than 160 specialized functions for structure prediction, drug design, and life science data analysis including genomics and proteomics. By connecting over 100 specialized databases and a large-scale knowledge graph, it supports the entire workflow: understanding research questions, selecting appropriate tools, predicting structures, analyzing results, and refining designs. The platform is designed to serve as an “AI Co-Scientist” assisting researchers.

Bae Choong-sik, President of KAIST, said, “National competitiveness in the AI era depends on ‘sovereign AI’ capabilities—the ability to develop and utilize core technologies independently. K-Fold is significant in that it combines South Korea’s proprietary AI technology with biotechnology to challenge world-class standards and connects it to services applicable in real drug discovery research.”

Bio AI has emerged as a critical technology for reducing the enormous time and cost of drug development, with big tech companies and research institutions in the United States, United Kingdom, and China engaged in intense competition for leadership. K-Fold is assessed as having laid the foundation for “sovereign bio AI” by securing core bio AI technology through South Korea’s own capabilities rather than relying solely on foreign models.

The achievement was first unveiled on June 23 at the “2026 Korean Society for Molecular and Cellular Biology Joint Academic Conference,” where Professor Kim Woo-youn delivered a keynote address titled “Generative Drug Design Powered by Agentic AI.” He stated, “K-Fold was developed not to follow existing models but to apply a new AI architecture that overcomes the limitations of conventional approaches. We will develop it into a scientific AI platform that makes world-class bio AI accessible to all researchers.”

Team KAIST plans to distribute K-Fold free of charge. HyperLab will provide beta services to domestic and international researchers before gradually expanding commercial services by the end of the year. The Korea Pharmaceutical and Bio-Pharma Manufacturers Association and the Korea Biotechnology Industry Organization will oversee the dissemination of K-Fold’s achievements and promotion of industry adoption.