Krafton-SKT Select Elite Team for Proprietary AI Foundation Model Project

Krafton, in collaboration with the SKT consortium, spearheads the development of a Korean-style AI foundation model. Krafton has emerged as a key player poised to lead next-generation AI technology, having been selected as an elite team for the Ministry of Science and ICT's national core project, the development of a 'Proprietary AI Foundation Model,' as a core member of the SK Telecom consortium. In this project, the SK Telecom consortium aims to build a proprietary full-stack AI encompassing semiconductors, models, data, and services, and to enhance accessibility by open-sourcing it to the domestic AI ecosystem. Specifically, the strategy is to implement Korean-style AI services (K-AI) through post-transformer AI models, thereby securing a leading position in the global AI market. Within this consortium, Krafton will play a core role by leading the architectural design and learning algorithm research for next-generation multimodal models. Based on this, it plans to develop AI solutions optimized for various domains, including games, and accelerate the development of APIs that can be applied to innovative content such as in-game AI NPCs and story engines. Krafton's unique strength lies in its proprietary platform that organically collects and utilizes multimodal data such as Vision, Text, Speech, and Action from actual game-play data. This provides overwhelming competitiveness in terms of securing core technologies and practical industrial application and is expected to make a critical contribution to the development of a proprietary Korean AI foundation model. Krafton has already demonstrated its Co-Playable Character (CPC) through collaboration with Nvidia and proven its technological prowess by unveiling 'Orak,' a benchmark for evaluating AI agents' game-play capabilities. It continues active research, having published 15 papers at leading AI conferences worldwide this year alone. Furthermore, it successfully applied a post-training technique co-developed with SK Telecom to three open-source inference-specialized language models with 7 billion parameters, once again confirming its capability in developing domain-specific AI models. Lee Kang-wook, Head of Krafton's Deep Learning Division, emphasized that "Krafton's unique technological prowess accumulated in the field of multimodal and game-specific AI will be a key differentiator for this consortium," and expressed his ambition to "actively contribute not only to the development of the domestic AI ecosystem but also to securing global AI leadership through close collaboration with the SK Telecom consortium."