SK Telecom's elite team, which advanced to phase 2 of the government's "domestic AI foundation model" project, plans to apply multimodal functions in stages, it said on Thursday.

The elite team's "A.X K1" drew attention as a hyper-scale AI model of the 519B class, the first in South Korea to exceed 500 billion parameters.

SKT's elite team scored 9.2 out of 10 in an NIA benchmark evaluation, one of the assessment steps in phase 1 of the project. It tied for first place with LG AI Research among five elite teams. The NIA benchmark evaluation comprehensively assesses areas including mathematics, knowledge, long-form comprehension, reliability and safety.

A.X K1 was released under the "Apache 2.0" licence, and high openness is also cited as a feature. Under the licence terms, models developed this way can be used commercially and can be modified and redistributed.

Starting from the phase 2 evaluation, SKT's elite team will apply multimodal functions to A.X K1 in stages, beginning with image data. From the second half of the year and beyond, it will upgrade multimodal capabilities so the model can also process voice and video data.

To improve AI model performance, the team will increase the scale of training data compared with phase 1. It will also expand the training languages to five (Korean, English, Chinese, Japanese and Spanish), accelerating development of a smarter AI model.

Cooperation and early-stage research at the elite team level are also active. The elite team is made up of eight organisations, including SKT, Krafton, 42dot, Rebellions, Liner, Selectstar, Seoul National University and KAIST. It recently broadened the scope of its research with the participation of the lab of KAIST Graduate School of AI Professor Minjun Seo and the lab of Seoul National University Department of Mathematical Sciences Professor Insu Seo.

About 20 organisations, including SK Group affiliates such as SK Hynix, SK Innovation, SK AX and SK Broadband, as well as the Korea Foundation for Advanced Studies and the Chey Institute for Advanced Studies, will also gradually use the SKT elite team's model and are set to lead innovation in the domestic AI ecosystem.