SK Telecom passed the second-stage evaluation of the government's project for an "independent AI foundation model," winning recognition for its in-house AI model technology. The latest review also reflected factors such as the feasibility of applying the model to real services and its ripple effects on the ecosystem, which is seen as acknowledging SKT's service and infrastructure capabilities as well.

The Ministry of Science and ICT said on Aug. 18 the second-stage evaluation was conducted by combining 40 points for benchmarks, 35 points for expert review and 25 points for user review. It increased the weighting for assessments of on-the-ground AI transformation (AX) integration and expansion compared with the first-stage evaluation, and it added a user review in which the general public directly used the model.

The expert review assessed not only model performance but also usability and practicality, applicability across industries and services, and ripple effects on South Korea's AI ecosystem. It in particular allocated 15 points to ripple effects and contribution plans, examining plans to support AX across fields, disclose AI models, and support universities, startups and the domestic AI semiconductor ecosystem.

SKT received high marks in this area. The expert committee assessed that SKT's A.X K2 secured top-tier performance among comparison models in mathematical reasoning and Korean. It also rated highly that the model is installed in large commercial services and showed a clear edge in usability and practicality.

The scope of industrial application was also cited as a strength. SKT provided the A.X K1 model to the defence sector and demonstrated a manufacturing-industry specialised agent. It also showcased cases of model use in legal and tax services. The expert committee assessed that these showed SKT proved applicability and practicality of its developed models across a range of industries.

It also stood out in the benchmark evaluation. The A.X K2, with 688 billion parameters, recorded top-level performance among global open-source models in Korean and mathematics benchmarks. In an evaluation of 2026 International Mathematical Olympiad problems, it posted a score corresponding to the gold medal standard. It also recorded a shared top score of 97.1 percent on the Math Arena AIME 2026 leaderboard based on problems from the U.S. high school maths competition AIME.

SKT will actively use government support to focus on expanding its in-house AI model into industry and everyday life. The ministry plans to raise the number of B200 GPUs provided to elite teams advancing to the next stage to about 1,000 in the second half from about 768 in the first half of this year.

An SKT official said, "Based on the results of expanding our in-house AI model across industry and everyday life, we will focus on producing outcomes that the domestic AI ecosystem can use together."