PANGYO, Korea (September 26, 2024) – NCSOFT today introduced the ‘Llama-VARCO LLM’, a highly efficient large language model (LLM) built on an open-source foundation.

‘Llama-VARCO LLM’ is built on the open-source model, ‘Llama 3.1’, advanced with NCSOFT’s proprietary technology, enhancing its Korean-language processing capabilities. It stands out for its compact size and top performance, the model ranks highest in the ‘Logickor’ benchmark, designed to assess reasoning capabilities in Korean-based LLMs, outperforming all other models under 10 billion parameters.

Engineered for broad compatibility, ‘Llama-VARCO LLM’ excels in diverse applications such as Chatbots, QA (Quality Assurance), and natural language generation. Its compact, high performing architecture provides an optimal solution for companies seeking efficient LLMs tailored to Korean-language AI services.

As the first Korean game company to develop its own proprietary language model, NCSOFT continues to deliver innovative and cost-effective solutions to meet market demands. In April 2024, NCSOFT launched the advanced multilingual model, ‘VARCO LLM 2.0’, followed by the recent introduction of ‘VARCO Judge LLM’, a specialized evaluation model for measuring LLM performance.

To promote AI ecosystem growth and support open research initiatives, NCSOFT is releasing ‘Llama-VARCO LLM’ as open-source on the AI platform, Hugging Face. This reflects NCSOFT’s ongoing commitment to fostering global collaboration and advancing AI technology through open-source contributions.

For further details, please visit the official NC Research website.