SK Telecom logo. /Courtesy of the company

SK Telecom said on the 16th that it advanced to phase two of the government's "independent AI foundation model project."

The elite team at SK Telecom drew attention with "A.X K1," a hyper-scale AI model with more than 500 billion parameters. A.X K1 excels in math (AIME25 benchmark) and coding (LiveCodeBench) and showed performance on par with or better than the global open-source model DeepSeek-V3.1.

In phase-one NIA benchmark evaluations, SKT's elite team scored 9.2 out of 10 to tie for first place with LG AI Research Institute. The NIA benchmark evaluation comprehensively assesses areas such as math, knowledge, long-form understanding, reliability and safety.

A.X K1 is released under the Apache 2.0 license, allowing commercial use as well as model modification and redistribution. In phase two, the team plans to apply multimodal capabilities starting with image data, followed by advancing the ability to process voice and video data.

To boost the AI model's performance, the elite team will expand the scale of training data and extend training languages to five languages (Korean, English, Chinese, Japanese and Spanish). This research, joined by eight institutions including SKT, Krafton, 42dot, Rebellions, Seoul National University and KAIST, is expected to drive innovation in Korea's artificial intelligence ecosystem.

※ This article has been translated by AI. Share your feedback here.