*This content was translated by AI.
KT announced on the 5th that its own super-large AI model 'Trust: Um K' has ranked first among small and medium-sized models in Korea in the 'Artificial Analysis Intelligence Index' (AAII), a platform that comprehensively evaluates global AI model performance.
AAII is a global evaluation platform operated by Artificial Analysis, an AI evaluation agency. We provide objective and reliable AI model information to users by combining the performance of major AI models around the world with multiple public benchmark results rather than a single test. Major AI models in Korea, such as LG Exem, Naver Hyperclova, Upstage Solar, and Motif, are also listed to promote Korea's AI competitiveness.
In this evaluation, M K scored excellent overall in more than 10 key evaluation items, including reasoning, expertise, mathematics, programming, and agent performance. Regarding this, KT emphasized that Believing:MK objectively proved that it has global competitiveness as an agent AI that does not only respond to users' questions but also understands the goals and then uses the necessary systems or tools to complete tasks on its own.
In this AAII release, Believing:MK is listed as the top model in the field of Agnostic AI, hitting 87% on one of the key agent performance benchmarks, the τ²-bench. τ²-bench is an agent-specific benchmark that evaluates whether AI can collaborate with humans in real-world work environments and use multiple tools to perform tasks to the end. KT said that the figures recorded by Trust: Um K are at the highest level and are on par with Google's latest model, the Gemini 3.
Belief: In addition to the agentic metrics, MMLU Pro, GPQA, HLE, and other key benchmarks that evaluate expertise and high-level reasoning skills have evenly outperformed. In particular, it achieved the highest level in Korea in terms of Korean comprehension performance indicators. These achievements are all the more remarkable in that they are implemented in sub-parameter 40B (40 billion) models. While increasing efficiency by reducing the burden of large-scale GPU infrastructure, it has secured the highest level of performance, including "instruction execution ability" to accurately understand users' complex instructions and "Korea-specific performance" to fully understand Korea's unique context and social context.
KT said it has created the most Korean AI optimized for the Korean business environment by implementing "Believe: Um K" in a "from scratch" method that directly performs the entire process from model architecture design to learning data construction. Believe it: Um K is designed to respond to a variety of scenarios that are immediately available in the field, including reporting, document summarization, data analysis, and customer response.
KT is planning to expand Trust: Um K into an A-type AI model optimized for the domestic B2B market. The strategy is to lead the AI transition (AX) of corporate customers by providing customized AI agents that automate and judge and process actual tasks of companies by industry such as finance, public, and manufacturing.
Oh Seung-pil, head of KT's technology innovation division (vice president), said, "This AAII registration is an achievement that objectively proves that KT's own AI technology has reached global standards," adding, "We will become the best AI partner to lead Korean companies' work innovation and digital transformation through AI."
<© STARNEWS. All rights reserved. No reproduction or redistribution allowed.>
*This content was translated by AI.
