LG Uplus said on November 15 that it will enter the business-to-business market next year with token-optimization technology designed to reduce the operating costs of AI agents. Its approach combines proprietary models, model routing, workflow controls and mixed GPU-NPU infrastructure. The company reported about 90% savings on coding costs using an in-house model, while model routing reduced general-inquiry costs by more than 25% and coding-agent costs by more than 30%. LG Uplus also reduced on-device model sizes from 2.4–3 gigabytes to about 800 megabytes while maintaining more than 95% of cloud-model performance, and said speculative decoding doubled token-generation speed. The company has used NVIDIA and AMD GPUs and worked with FuriosaAI for more than two years to validate NPUs. Earlier disclosures cited additional gains from model compression and harness engineering, alongside plans to build AI-transformation experience through Everyone's AI with Kakao.