LG CNS and Canadian AI Startup Unveil Advanced Inference LLM... Foundation for Agentic AI Services Established

Joint Development of 111 Billion-Parameter LLM with Cohere
Operable with Only Two GPUs... Aiming to Secure 'Sovereign AI' Competitiveness

LG CNS Headquarters Exterior. Photo by LG CNS

On July 10, LG CNS announced that it has jointly developed a large-scale inference-type large language model (LLM) with 111 billion parameters in collaboration with the Canadian artificial intelligence (AI) startup Cohere. This model is a core technology for realizing 'agentic AI', where AI makes its own decisions and performs tasks autonomously, and supports 23 languages.

According to internal testing by both companies, the newly developed LLM demonstrated superior inference capabilities in Korean and English compared to leading international models such as ChatGPT-4o and Claude 3.7 Sonnet. In particular, it achieved higher scores than other models on the Math500 and AIME 2024 benchmarks, which evaluate mathematical reasoning abilities. The model also supports an on-premise approach, allowing sensitive data to be processed securely within internal infrastructure. Cost efficiency has been enhanced through compression technology that enables operation with just two graphics processing units (GPUs). With this, LG CNS aims to secure 'sovereign AI' competitiveness suitable for sectors such as public services and finance.