Cost-Effective Models Hit the Market... Competition for Token Efficiency Heats Up
'Model Routing' Gains Ground, Letting Companies Choose Models by Task
SK Telecom Boosts Productivity by Designing an Agent-Based Organization

The AI competition is shifting from model performance to efficiency. As global tech giants introduce low-cost, high-efficiency models, companies are also changing their work practices and organizational structures to boost productivity through AI.

AI Enters the 'Cost-Effectiveness' Era... Corporate AX Changes Amid Efficiency Race View original image

According to industry sources on October 29, companies adopting AI are making it a key priority to improve results relative to the number of tokens used. OpenAI has launched GPT-6.1 Sol, which offers performance close to its top-tier models at one-fifth the price, while Anthropic has introduced Claude Opus 5.5, with token prices cut by 20%.


“Model routing,” which uses lightweight and high-performance models according to the complexity and importance of a task, is also gaining ground. Anthropic has organized Claude into tiers: Haiku, a low-cost lightweight model; Sonnet, a mid-tier model; Opus, a high-performance model; and Fable, an ultra-high-performance model. Amazon Web Services (AWS) Bedrock provides enterprise infrastructure that lets users choose from more than 100 AI models on a single platform based on the type of work.


There is also a growing trend toward charging based on actual deliverables rather than AI usage. U.S.-based AI document platform Mintlify has switched to a pricing structure based on deliverables rather than token usage. India’s voice AI sales company SquadStack has also introduced a model that ties fees to actual revenue generated rather than the number of customer interactions handled.


Companies in South Korea are also considering “token cost-effectiveness” as one of their performance indicators. At the Samsung AI Forum 2026 in September, Samsung Electronics stressed that the focus of AI development should shift from scaling up to efficiency. For example, its hybrid AI, used in Galaxy AI, handles tasks requiring a quick response or privacy protection on the device, while processing only tasks that require large-scale computing in the cloud.


SK Telecom is also viewing AI not merely as an assistant tool but as an agent that performs work, with the aim of improving work efficiency through AI. After running its One AI Agent per Person project in March, in which every SK Telecom employee personally created an AI agent and used it for work, the company is redesigning its work practices so that AI agents are assigned employee ID numbers and work alongside staff as members of the organization.


SK Telecom CEO Jeong Jaeheon presented the “AX J-Curve,” which applies economic theory to the AI transformation process, during a special lecture at Seoul National University in September. He explained that productivity may decline in the early stages of AI adoption but rise after that initial phase, and that redesigning work practices is essential to achieve a sustained productivity increase.



An industry official said, “Companies that establish systems for managing token efficiency in the early phase of the J-curve and turn token consumption into tangible business value will take the lead in the AI era.”


This content was produced with the assistance of AI translation services.

© The Asia Business Daily. All rights reserved. Unauthorized AI training and use prohibited.

Today’s Briefing