MOBILUS

Motif Technologies has been selected as an elite team for South Korea’s government-led Independent AI Foundation Model project. The company plans to develop a large-scale multimodal foundation model covering text, image, video, and audio, advancing toward 300B-parameter reasoning LLMs, 310B VLMs, and 320B VLA models. Motif emphasized its fully independent architecture—introducing innovations such as Group-wise Differential Attention (GDA) and PolyNorm GLU—rather than adopting existing foreign open-source transformer structures. The company also aims to open-source model weights, code, and optimization libraries to foster a domestic AI ecosystem and accelerate AI transformation (AX) across industries and the public sector.
The consortium includes 17 participating and 12 demand-side organizations spanning model development, infrastructure, data, and real-world deployment. Motif will lead model design, while its parent company Moreh will handle large-scale GPU optimization and distributed inference. Academic partners such as KAIST, Seoul National University, and Hanyang University will support multimodal research and data technologies. In industrial applications, projects will span smart homes, education, cultural heritage, robotics, and autonomous systems. Notably, Mobilus will conduct demonstrations of physical AI–based autonomous driving and operations in agricultural and off-road environments. With the selection, Motif will receive substantial GPU and data support, and a second-stage evaluation is scheduled for early August.

Link: https://www.aitimes.com/news/articleView.html?idxno=207025