본문 바로가기
bar_progress

Text Size

Close

SKT Elite Team to Sequentially Apply Multimodal Capabilities in Phase 2 of Independent AI Foundation Model Project

Korea's First 500B-Scale Model "A.X K1"
Advanced Mathematical and Coding Reasoning, High Openness
Expanded Training Data and Added Multimodal Capabilities

SK Telecom's elite team announced on January 16 that, starting with image data, they plan to sequentially apply multimodal capabilities in the second phase of the government's "Independent Artificial Intelligence (AI) Foundation Model Project."


SKT Elite Team to Sequentially Apply Multimodal Capabilities in Phase 2 of Independent AI Foundation Model Project The 'A.X K1' presented by SK Telecom's elite team. Provided by SK Telecom


The "A.X K1" presented by SK Telecom's elite team is the first ultra-large AI model in Korea to surpass 500 billion parameters, reaching a scale of 519 billion. A.X K1 demonstrated strong performance in advanced mathematics and coding domains. In areas such as mathematics and coding applications, it showed performance comparable to or better than global open-source models with similar parameter sizes, such as DeepSeek-V3.1. In addition, A.X K1 has been released under the Apache 2.0 license, which allows for commercial use and permits modification and redistribution of the model according to licensing regulations.


In the first-phase benchmark evaluation by the National Information Society Agency (NIA), A.X K1 scored 9.2 out of 10, tying for first place with LG AI Research among the five elite teams. The NIA benchmark comprehensively evaluates various areas, including mathematics, knowledge, long-form understanding, reliability, and safety.


Starting from the second-phase evaluation, SK Telecom's elite team plans to sequentially apply multimodal capabilities, beginning with image data. This will enable A.X K1 to recognize images from academic papers or work documents and summarize them into text. In the second half of this year, they will further advance multimodal capabilities to process voice and video data as well. To achieve this, they will expand the scale of training data compared to the first phase and increase the number of training languages to five (Korean, English, Chinese, Japanese, and Spanish), accelerating the development of an even more advanced AI model.


An SK Telecom representative stated, "Recently, the research scope has expanded with the participation of Professor Seo Minjun's AI Graduate School lab at KAIST and Professor Seo Inseok's Department of Mathematical Sciences lab at Seoul National University," adding, "More than 20 organizations, including SK Group affiliates such as SK Hynix and SK Innovation, as well as the Korea Foundation for Advanced Studies and the Choi Jonghyun Academy, will gradually utilize the SK Telecom elite team's model, leading innovation in Korea's AI ecosystem."


© The Asia Business Daily(www.asiae.co.kr). All rights reserved.


Join us on social!

Top