KT said on Sept. 27 that its in-house AI model routing technology, 'Auto Model Router', ranked second in a global public benchmark.

Auto Model Router placed second overall on 'Router Arena', a large language model (LLM) router evaluation platform developed by researchers at Rice University in the United States.

Router Arena is a public benchmark that evaluates AI routers based on about 8,400 queries. It assesses response accuracy, cost efficiency and robustness to input changes. It currently lists both academic routers and commercial technologies, including Microsoft's 'Azure Model Router'.

Auto Model Router automatically selects the model best suited to a user's request from multiple AI models. It analyses the type and difficulty of the task and the knowledge domain, and links the request to a model after considering response quality and usage costs by model. For example, it links cost-efficient models for translation or simple information checks, and high-performance models for specialised analysis or tasks requiring advanced reasoning.

KT also uses Auto Model Router for the model routing function of its 'Token Factory'. As it integrates and operates multiple AI models and token-usage environments, it automatically selects the model that fits each request to optimise both service quality and cost.

KT plans to further develop Auto Model Router and build a multi-model operating environment that can add new AI models.

Jun-seok Kim (김준석), an executive director and head of KT's Agentic AI Lab, said, "In the AI era, it is important how intelligently you use the optimal model for each situation." He added, "Auto Model Router will become a core technology that supports the competitiveness of KT's agentic AI services, including Token Factory."