KT’s AI Model-Routing Technology Ranks Second in Global Benchmark
KT says its internally developed Auto Model Router placed second in a global benchmark that measures how well AI routing systems balance response accuracy and operating cost.
South Korean telecommunications company KT said Sunday that its internally developed artificial intelligence model-routing technology ranked second in a global public benchmark, highlighting the company’s effort to improve the economics and performance of enterprise AI services.
The technology, called Auto Model Router, placed second in the overall category of RouterArena, an evaluation platform developed by researchers at Rice University in the United States. According to KT, the category assesses both accuracy and cost rather than rewarding systems solely for selecting the most capable or least expensive large language model.
AI routing systems choose which model should handle a particular user request when several models are available. A router may analyze the type of task, its difficulty and the relevant field of knowledge before directing the request to a model. The selection process can also account for differences in response quality and usage fees among models.
That function is becoming increasingly relevant for companies deploying AI at scale. Sending every request to a high-performance model can increase token costs, while automatically choosing the cheapest model can reduce output quality. Routing technology is designed to manage that trade-off by matching each request with a model under a defined policy.
KT said Auto Model Router received a strong evaluation because it applies a selection policy intended to meet a company’s target quality level while improving cost efficiency. The approach does not uniformly choose either the strongest or the least expensive model. Instead, it aims to make a context-sensitive choice based on the request and the customer’s operational priorities.
The company plans to use the technology as part of the model-routing function in Token Factory, an AI initiative it is developing. KT said the integration is intended to optimize service quality and token-use costs at the same time.
Kim Jun-seok, an executive at KT’s Agentic AI Lab, described Auto Model Router as a core technology for implementing AI orchestration. He said it would support the competitiveness of KT’s agentic AI services, including Token Factory, and help the company lead enterprise customers’ AI transformation.
The benchmark result gives KT a public measure of how its routing approach compares with other systems on a combined quality-and-cost basis. Its commercial impact, however, will depend on how effectively the technology performs when applied to enterprise workloads and customer requirements.