KAIST researchers have unveiled K-Fold, a new foundation model for molecular-complex structure prediction and drug-design work. The team says the system makes South Korea the fourth country after the United Kingdom, United States and China to hold an independent co-folding capability at this level. K-Fold removes the multiple-sequence-alignment preprocessing step used by many existing systems and reports inference speeds up to 25 times faster in some workflows.
Researchers also reported stronger results than AlphaFold 3 on selected GPCR and kinase benchmarks, along with an improvement in target-protein-degradation prediction success from about 19 percent to 25 percent. The 7B and 2B models are being released under the Apache 2.0 license, while a web platform integrating K-Fold is expected to launch commercially this year. The practical significance is access: Korean labs and biotech companies can test a domestic model without relying entirely on closed foreign systems.