Artificial intelligence startups in South Korea are rolling out services that cut AI token costs, targeting the business-to-business market. As corporate AI transformation spreads and the use of large language models and AI agents grows sharply, more companies want to run their AI budgets more efficiently.
Liner launched an AI orchestration solution called Liner Model API on the 15th, according to venture industry sources on the 16th. The service routes difficult queries such as coding, mathematics and complex reasoning to high-performance frontier models, while handling ordinary queries with relatively inexpensive lightweight models to reduce token costs.
The Liner Model API is a commercialized version of technology the company originally developed in-house to cut costs while running its consumer AI service. An analysis of its service data showed that only about 43% of all queries actually required a high-performance model, the company said. By assigning the optimal model to each query, it says token costs can be cut by more than half compared with routing every query to a frontier model, while maintaining frontier-level performance.
"Because this is technology we built in the first place for our own use, we expect strong demand across consumer AI companies operating on tight budgets," a Liner official said. Liner is currently testing the service with three startups, including Planfit and Vacatio, and plans to use them as references to expand its customer base.
"Starting with the API service, we will begin securing new growth drivers in earnest in the B2B area," Liner CEO Kim Jin-woo said. "Our goal is to raise the B2B share of revenue to as much as 20% within the year, completing a more balanced earnings structure."
Wrtn Technologies has also moved to cut token costs with Wrtn AX LLM Router, a corporate service that automatically selects the most suitable LLM based on the intended use and on whether cost or performance takes priority. Wrtn AX, an in-house company under Wrtn, launched the service in July. That same month, Wrtn provided product credits to Cold Fusion, a team that advanced to the finals of the international hacking competition DEF CON 34 CTF, securing a use case.
"When customers actually adopt it, we build the router function into their on-premises or private cloud environment," Wrtn AX CEO Park Min-jun said. "We have confirmed demand across a range of sectors, including manufacturing, the public sector and finance."
Some startups control corporate token spending before it happens. Boundfor, a data foundry company, last month added a token spending management service called Pareto to Dropi, its AI data operations platform. It blocks AI API requests that exceed a budget set by an administrator before they are executed, and shows usage and projected spending by team in real time. It also identifies where costs are leaking, such as repeated or duplicate calls and unnecessary use of high-performance models.
"Apart from adopting AI, building a system to review and control costs afterward also takes staff and time," Boundfor CEO Hwang In-ho said. "So our goal is to reduce the AI transformation burden for non-IT companies that lack in-house development staff."
