KT's self-developed AI model routing technology, 'AutoModelRouter,' recorded a comprehensive second-place ranking in the global benchmark for LLM (large language model) router categories.
KT announced on the 27th that its AutoModelRouter achieved second place on the 'Router Arena' public leaderboard based on the 'Acc-Cost Arena' (accuracy-cost evaluation) criteria. Router Arena is an LLM router evaluation platform developed by researchers at Rice University in the United States, which comprehensively evaluates AI router performance using approximately 8,400 queries, including response accuracy, cost efficiency, and robustness against input variations.
AutoModelRouter is a technology that automatically selects the most suitable model among multiple AI models based on user requests. It analyzes the task type, difficulty level, and knowledge domain of the request, then connects to a model by considering the response quality and usage costs for each model.
The system is designed not to simply select high-performance or lowest-cost models, but to choose models that meet target quality requirements set by enterprises while enhancing cost efficiency. For instance, it connects cost-effective models for simple translation or information verification tasks, and high-performance models for tasks requiring advanced reasoning.
AutoModelRouter is also utilized in the 'Token Factory' model routing function being promoted by KT. As Token Factory integrates and operates various AI models and token usage environments, AutoModelRouter automatically selects a model that matches user requests.
Kim Jun-seok, Senior Executive and Head of KT's Agentic AI Lab, stated, "KT will lead the AI transformation of corporate customers based on its competitive AI platform technologies."
in TECH
