KT, self-developed AI model routing technology ‘Auto Model Router’ ranked 2nd overall in public benchmark

Find and select a model that suits the user’s request… Also used in token factory

(Seoul = Yonhap News) Reporter Kim Joo-hwan = KT [030200] announced on the 27th that its self-developed AI model routing technology 'Auto Model Router' ranked second overall in a public benchmark specialized in large language model (LLM) routers.

According to KT, Auto Model Router ranked second in the comprehensive category that evaluates both accuracy and cost in the public leaderboard of 'Router Arena', an LLM router evaluation platform developed by researchers at Rice University in the United States.

Routing is a technology that automatically selects the model that best suits the user's request among several AI models.

It is important to have the ability to analyze the ▲ task type ▲ difficulty ▲ knowledge field of the request and connect the model that will handle the request according to a routing policy that takes into account the response quality and usage cost for each model.

In particular, KT received high praise for applying a model selection policy to increase cost efficiency while meeting the target quality required by the company, rather than simply uniformly selecting the highest performing model or the cheapest model.

KT plans to use Auto Model Router for the model routing function of 'Token Factory', which KT is promoting, to optimize the quality of AI services and the cost of token use.

Kim Jun-seok, managing director of KT Agentic AI Lab, said, "Auto Model Router is a technology that implements these AI orchestration capabilities, and will be a core technology that supports the competitiveness of KT's agentic AI services, including Token Factory. We will lead the AX transition of corporate customers based on our competitive AI platform technology."