SKT Vice President Taeyoon Kim: "Supplying Quantized A.X K2 to Ministry of National Defense"
"Expanding into Public and Industrial Domains Through Enhanced Security"
"Plans to Scale Up the Next Model to Trillion-Parameter Level"
As the second round of evaluations for the “Independent AI Foundation Model” (Dokpamo), which selects Korea’s national representative artificial intelligence (AI), has begun, SK Telecom unveiled its enhanced “A.X K2” model, which features improved performance and inference efficiency. This new model consists of 688 billion parameters and is distinguished by its increased average performance across 14 domestic and international benchmarks.
Taeyoon Kim, Vice President in charge of Foundation Models at SK Telecom, announced through the newsroom on August 4 that the “A.X K2” model’s average performance across 14 domestic and international benchmarks improved by 32.2 percentage points compared to the previous model, “A.X K1,” with advancements of approximately 83.9 percentage points in long-form comprehension and agent-related evaluations.
In particular, the “A.X K2” further increased corporate utility by systematizing tacit knowledge from manufacturing sites. Kim stated, “Companies utilizing A.X K2 can convert the tacit knowledge held by skilled workers, as well as scattered field documents, into digital assets—enabling better quality judgment and greater production efficiency.” To achieve this, SK Telecom’s elite team independently developed Korean-language manufacturing AI benchmarks for the semiconductor, automotive, and chemical sectors and established a system seamlessly connecting data collection, training, and evaluation.
Kim emphasized that, from a security perspective, the on-premises deployment history has provided SK Telecom with the capability to expand into public and industrial domains. Utilization cases linked to public safety, such as external phishing detection bots and real-time voice phishing detection based on audio language models, are also being broadened.
Kim stated, “For security-critical areas such as defense, we are supporting model quantization to reduce memory and processing loads, allowing stable operation even in resource-constrained environments,” adding, “By the second half of this year, we plan to provide a quantized version of A.X K2 to the Ministry of National Defense.”
SK Telecom plans to scale up the next-generation model succeeding A.X K2, increasing the parameter count to the trillion-unit range. Kim said, “We will further advance agentic reinforcement learning, cybersecurity capabilities, efficiency in long-context processing, and fast serving technology,” and added, “While conducting next-generation architecture research in collaboration with Seoul National University, KAIST, and Krafton, we also plan to strengthen optimization for model serving and commercial service validation based on domestically developed neural processing units (NPUs) in conjunction with consortium partner Rebellions.”
Hot Picks Today
"Two Spend 500,000 Won in a Flash, Yet Wait Two Hours?" ... 'Dreamland' Shunned by Kids and Foreign Tourists
- The Era of Getting 200 Million Won After 30 Years Is Over... A Japanese Bank's "Perpetual Department Head" Who Just Clocked In Cries Out
- Spain Hit by Its Worst Tragedy... Lee Kangin, Moreno and Real Madrid Join Soccer World in Mourning
- They Said, “Make It Your Home for Life” - Then Suddenly Kicked 200 Passengers Off... What Happened to the Around-the-World Cruise?
- "130 Rice Bowls Sold Out in a Day"... 'Unfiltered Face' Adds 8 Million Won to Night Market Sales: China's Bungeunwoo
Kim added, “We will continue our free licensing policy by distributing the models developed through the Dokpamo project under the Apache 2.0 License, enabling developers and startups to use them for commercial purposes.”
© The Asia Business Daily. All rights reserved. Unauthorized AI training and use prohibited.