KT’s AutoModelRouter takes second place on RouterArena
KT says its AutoModelRouter placed second on RouterArena’s ranking that combines response accuracy and cost. The entry, listed as KT-ModelRouter, stands behind Paix2 on the public leaderboard. The system routes a user request to a language model according to its task, difficulty and knowledge domain. The ranking shows benchmark performance; it does not establish savings in customer deployments.
Artificial Intelligence··Midday
KT-ModelRouter places second on accuracy and cost
KT said on September 27 that AutoModelRouter placed second in RouterArena, a benchmark developed by Rice University researchers. ChosunBiz and Dailian identify the entry on the public leaderboard as KT-ModelRouter and specify that the placing is in Acc-Cost Arena, which considers answer accuracy together with cost. It is a ranking of systems that route requests among language models under a particular measure, not a general ranking of all language models. The public list shows KT-ModelRouter at 76.28, behind Paix2 at 77.63.[1], [2]
The router selects a model by request type
AutoModelRouter is meant to choose among language models for each request rather than send every user to one model. According to Dailian, it evaluates the task type, difficulty and knowledge domain while weighing answer quality and processing cost. The leaderboard includes academic routing systems as well as commercial offerings such as Azure Model Router. KT’s second place therefore concerns the routing decision under the benchmark, not the popularity of a chat interface or the output of a single model on every question.[2]
The benchmark uses about 8,400 queries
RouterArena evaluates routing systems on about 8,400 queries, according to ChosunBiz, and the paper behind the benchmark was accepted at ICLR 2026. That scope matters to the meaning of the placing: the score comes from a defined query set and an accuracy-and-cost measure. The coverage does not measure how widely KT customers use the router in production or what savings they obtain from it. The disclosed development is the system’s standing on this public benchmark.[1]
