KT is set to unveil a platform that allows users to access more than 500 global AI (artificial intelligence) models, including ChatGPT, in one place while also managing costs. The core features include automatically selecting the AI model best suited for specific tasks and compressing input data to reduce token usage by up to 25%.
On the 27th, KT announced that it would reveal its AI platform, 'KT Token Factory,' at the Seoul Jongno-gu Gwanghwamun West Building and launch it on the 30th. Initially, the service will be provided to corporate clients, with plans to expand the target user base in the future.
Token Factory is a service that supports the use of multiple AI models through a single platform. Companies can handle everything from model selection to usage monitoring, cost management, and payment in an integrated manner, without needing to create separate accounts or sign individual contracts for each AI model.
As AI models have become more diverse, companies are utilizing different models for various tasks. This is because the suitable model varies depending on the work, such as coding, translation, document summarization, and complex reasoning. However, subscribing to multiple models individually makes it difficult to manage usage status and costs.
KT addresses this issue with its self-developed 'Auto Model Router.' This technology analyzes the type of user query, difficulty level, required quality, and cost to automatically select the appropriate AI model. It assigns low-cost models for simple tasks and high-performance models for complex analysis.
This technology ranked first overall in 'RouterArena,' a global benchmark that evaluates AI model selection performance. KT also applied prompt compression technology that considers the characteristics of the Korean language. This reduces the number of input tokens by removing unnecessary expressions while preserving the meaning of the query. According to the company, these two technologies can reduce token usage by up to 25%.
Data security has also been strengthened. KT provides inference services for its own AI models and open-weight (models with public weights) AI models from domestic data centers. This is done to alleviate concerns about corporate data being transmitted to overseas servers. The company has also applied its proprietary guardrail technology, which detects personal information, sensitive data, and harmful expressions.
KT signed an official partnership with the global AI model distribution platform 'OpenRouter,' marking the first such agreement in South Korea. Through this partnership, KT has secured more than 500 global AI models and plans to promote the overseas distribution of AI models provided by KT within South Korea. The company aims to expand Token Factory from a corporate AI platform into a global AI model distribution business. It is collaborating with domestic and international companies such as Eucal, CAST AI, and Timely, and is also building an ecosystem with participants in the 'AI for All' consortium.
Park Sang-won, Senior Executive Vice President and Head of KT's AX Business Division, stated, "The AI market is rapidly shifting from a competition over who can build better models to one over who can support the easier and more efficient utilization of AI." He added, "We will develop Token Factory into a core platform that drives the growth of South Korea's AI ecosystem."
in TECH
