Motif-Technologies Releases Motif 3 Beta, a 314B Parameter MoE Model
Motif-Technologies has released the beta version of Motif 3, a 314-billion parameter Mixture of Experts (MoE) model. This model was developed as part of South Korea's state-funded "Independent AI Foundation Model" project. The release highlights South Korea's highly competitive, state-funded initiative to build open-source foundation models, aiming to drive domestic AI innovation and reduce reliance on foreign technology. The survival-of-the-fittest funding structure forces participating companies to rapidly iterate and deliver high-performing models. Motif 3 Beta features 314 billion total parameters with 13 billion active parameters (314B-A13B), positioning it alongside other MoE models like DeepSeek V4 Pro and MiniMax-M3. Motif Technologies was added to the project in February 2026 after Naver Cloud and NC AI were eliminated during the first evaluation in December 2025.
## BACKGROUND
South Korea's AI Foundation Model project is a government initiative investing ₩530 billion ($0.36 billion) until 2027 to support domestic tech companies in developing open-source AI models. Mixture of Experts (MoE) is an architecture that uses conditional computation, loading only a subset of parameters (active parameters) for each token to scale model capacity efficiently.