[Edaily Reporter Han Kwangbeom ] KTCorporation(030200)announced on the 27th that its proprietary AI model routing technology, “AutoModelRouter,” has proven its technological competitiveness by ranking second overall on “RouterArena,” a public benchmark specializing in large language model (LLM) routers.

RouterArena is a router evaluation platform developed by researchers at Rice University in the U.S., and the related research was accepted as a full paper at ICLR 2026, an international conference in the field of machine learning. It comprehensively evaluates performance metrics essential for real-world service operations—such as response accuracy, cost efficiency, and robustness to input variations—using approximately 8,400 queries.

AutoModelRouter was listed on RouterArena’s public leaderboard under the name “KTCorporation-ModelRouter” and ranked second in the metric that evaluates both accuracy and cost. Through this, it demonstrated its technological competitiveness in optimizing both quality and cost while selecting the most suitable model for a given use case from among various options. Currently, the public leaderboard lists not only routers developed by academia but also commercial technologies such as Microsoft’s (MS) Azure Model Router.

The numerous models currently available to businesses each have different strengths in areas such as translation, summarization, coding, and reasoning, and they also vary in terms of performance and usage costs. Therefore, rather than processing all tasks with a single model, the technology that selects the appropriate model based on task characteristics and required quality is emerging as a key element in the digital transformation (AX) process.

AutoModelRouter is a technology that automatically selects the model best suited to a user’s request from among multiple models. It analyzes the task type, difficulty, and domain of the request, and connects the request to the appropriate model based on routing policies that consider response quality and usage costs for each model.

Rather than simply selecting the highest-performing or lowest-cost model across the board, the system applies a model selection policy designed to meet the target quality required by the enterprise while also maximizing cost efficiency.

For example, cost-effective models that meet the required quality standards are used for relatively simple tasks such as translation or information verification, while high-performance models are deployed for tasks requiring specialized analysis or advanced reasoning. Although users interact with a single service, the system actually utilizes different models depending on the characteristics of each request.

This allows companies to reduce the burden of comparing and selecting multiple models one by one and to configure their operating environments more efficiently according to the specific requirements and budgets of each task.

AutoModelRouter is also utilized in the model routing functionality of the “Token Factory,” which is currently under development. As Token Factory integrates the operation of various models and token usage environments, AutoModelRouter plays a key role in optimizing both service quality and token usage costs by automatically selecting the model that best matches the user’s request.

Going forward, KT plans to continuously enhance the AutoModelRouter and build a multi-model operating environment that allows for the flexible addition of new models, thereby supporting enterprise customers’ adoption and digital transformation.

Kim Jun-seok, Head (Senior Vice President) of KTCorporation’s Agent-based AI Lab, stated, “The AutoModelRouter is a technology that embodies AI orchestration capabilities and will serve as a core technology underpinning the competitiveness of KTCorporation’s agent-based AI services, including Token Factory.” He added, “KTCorporation will lead the AX transformation for enterprise customers based on our competitive AI platform technology.”