Add preferred source
KT’s self-developed AI model routing technology, “AutoModelRouter,” has ranked second overall on RouterArena, an LLM router evaluation platform created by researchers at Rice University in the United States. The technology analyzes user requests and automatically selects and connects the most suitable AI model based on task type and difficulty. In an evaluation using approximately 8,400 queries, it placed second on the Acc-Cost Arena benchmark, which measures both accuracy and cost efficiency, competing against commercial technologies including Microsoft’s. KT plans to leverage this technology as a core feature of its AI infrastructure platform “Token Factory” to support enterprise customers’ AI transformation.
Key Elements
KT has developed technology that automatically selects and connects the most suitable AI model from multiple options based on a user’s question and task difficulty. The technology ranked second in a global public evaluation that comprehensively measured accuracy and cost efficiency.
On the 27th, KT announced that its self-developed AI model routing technology “AutoModelRouter” placed second overall on RouterArena, a large language model (LLM) router evaluation platform. It was listed on the public leaderboard under the name “KT-ModelRouter.”
An LLM router is a mechanism that analyzes a user’s request (query) and routes it to the most appropriate AI model. Since each AI model excels in different areas—such as translation, summarization, coding, and reasoning—and carries different usage costs, the router selects and connects the most suitable model for each request rather than having a single AI handle all tasks.
RouterArena is an LLM router evaluation platform developed by researchers at Rice University in the United States. It evaluates AI routers’ response accuracy, cost efficiency, and robustness against input variations using approximately 8,400 queries. The related research was accepted as a full paper at ICLR 2026, an international conference in the machine learning field.
KT’s AutoModelRouter ranked second on the “Acc-Cost Arena” benchmark, which evaluates accuracy and cost together. The leaderboard includes technologies developed in academia as well as commercial solutions such as Microsoft’s “Azure Model Router,” with which KT’s technology competes directly.
When a user request comes in, AutoModelRouter analyzes the task type, difficulty level, and knowledge domain. It then compares the response quality and usage cost of each AI model and automatically selects the model best suited to handle the task. For simple translation or information verification, it uses cost-efficient models, while for tasks requiring high-level reasoning or specialized analysis, it routes to high-performance models.
From an enterprise perspective, this reduces the need to manually compare which AI model to use for each task. Companies can build a multi-model environment that operates multiple AI models simultaneously, tailored to their required quality levels and budgets.
Integration with Token Factory
KT will also apply this technology to the model routing function of its proprietary AI infrastructure platform, “Token Factory.” As Token Factory integrates and manages multiple AI models and token usage environments, AutoModelRouter selects the appropriate model for each request, optimizing both service quality and token usage costs.
Enterprise customers can reduce the burden of directly comparing and selecting among multiple AI models, and build an optimal AI transformation (AX) environment aligned with their business requirements and budgets.
Kim Jun-seok, head of KT’s Agentic AI Lab (Senior Vice President), explained: “In the AI era, what matters is not just securing the most powerful single model, but how intelligently you leverage the optimal model for each situation. AutoModelRouter is a technology that embodies this AI orchestration capability.”
Future Plans
Going forward, KT plans to enhance AutoModelRouter’s performance and build a multi-model operating environment capable of incorporating new AI models. This is seen as a strategy to strengthen its competitiveness in the enterprise AI transformation market based on these evaluation results.
This achievement marks a case where AI infrastructure technology independently developed by a South Korean telecommunications company ranked among the top in a public evaluation involving global academia and corporations. It demonstrates the growing importance not only of AI model performance itself, but also of orchestration technologies that efficiently operate those models.
Once added, BigGo Finance appears first in Google Search Top Stories, so you get the broadest, most up-to-the-minute, and most comprehensive global financial news first.
