KT ranked second in a public benchmark of large language model (LLM) routers with its in-house technology “AutoModelRouter,” which automatically selects and consolidation the most suitable model among several based on the user’s artificial intelligence (AI) request.
KT said on the 27th that AutoModelRouter placed second under the “accuracy–expense evaluation (Acc-Cost Arena)” criteria, which assess both accuracy and cost efficiency, on “RouterArena,” an LLM router evaluation platform developed by researchers at Rice University. The public leaderboard lists it under the name “KT-ModelRouter.”
RouterArena is a platform that evaluates AI routers’ response accuracy, cost efficiency, and robustness to input changes using about 8,400 queries. The related research has been accepted as a full paper at the international Machine Learning conference “ICLR 2026.” The public leaderboard includes not only routers developed in academia but also commercial technologies such as Microsoft’s “Azure Model Router.”
AI model routing is a technology that automatically
AI model routing is a technology that automatically selects and consolidation the appropriate model among several AI models according to the user’s request. Because model performance and usage expense differ by task—such as translation, summarization, coding, and reasoning—requests are not all handled by a single model; instead, a model is chosen that fits the purpose and difficulty.
KT’s AutoModelRouter analyzes the user’s request by task type, difficulty, and knowledge domain, then determines which model to use by considering each model’s response quality and usage expense. For relatively simple translation or information checks, it consolidation a cost-efficient model that meets the required quality, while for tasks requiring professional analysis or a high level of reasoning, it consolidation a high-performance model.
KT said, “Instead of uniformly choosing the highest-performing model or the lowest-expense model, we designed the model selection method to meet the target quality corporations require while also improving cost efficiency.”
AutoModelRouter is also used in the model routing
AutoModelRouter is also used in the model routing function of the “Token Factory” that KT is promoting. In the integrated operation of multiple AI models and token usage environments, Token Factory automatically selects the model that matches user requests and manages both service quality and token usage expense.
KT plans to further advance AutoModelRouter and build a multimodel operating environment that can add new AI models to support corporations’ AI transformation (AX). Kim Jun-seok, head of KT’s Agentic AI Lab (senior vice president), said, “In the AI era, it is not only important to secure the single best model but also how intelligently you leverage the optimal model for the situation,” adding, “AutoModelRouter is a technology that implements this AI orchestration capability.”
※ This article has been translated by AI. Share your feedbackhere.
