LG CNS, Cohere co-develop Korean-specific agentic AI

LG CNS Co., the information technology services arm of South Korea’s LG Group, has unveiled a lightweight Korean-specific large language model (LLM) co-developed with Canadian AI startup Cohere Inc., aiming to deliver secure, cost-efficient AI solutions for Korean clients in the financial and public service sectors.

The newly released model is powered by 7 billion parameters, allowing for faster processing and lower operating costs than hyperscale global models backed by hundreds of billions of parameters, according to sources in the AI industry on Sunday.

Its compact architecture makes it well-suited for secure, on-premise deployments, a key requirement for institutions handling sensitive data.

The new model is expected to accelerate LG CNS’ push into the agentic AI market, especially in the financial and public service industries. 

Cohere, founded in 2019 by former Google AI researchers and backed by deep learning pioneer Geoffrey Hinton, is renowned for its enterprise-first AI models. Valued at roughly $5.5 billion, it serves major clients such as the Royal Bank of Canada, known for its stringent privacy and compliance standards.

In March, it agreed with LG CNS to develop industry-leading secure, customized agentic AI solutions together for Korean enterprises, marking the Canada-based startup’s first alliance with a Korean company to offer a Korean-language LLM.

AGENTIC AI

The partnership is expected to enable LG CNS to broaden LLM offers with flexible deployment options, including both cloud-based and in-house installations, critical for regulated sectors like banking and government.

In benchmarking tests, the LG CNS-Cohere model outperformed five open-source Korean LLMs of similar size across 25 evaluation categories, highlighting its competitive performance despite a more compact design.

The launch builds on LG CNS’ strategic push into agentic AI, which not only generates content but also autonomously plans, reasons and executes tasks across platforms.

Unlike generative AI, agentic models are capable of navigating complex workflows, making decisions and interacting with enterprise systems with minimal human supervision.

With many Korean financial service providers investing as much as 100 billion won ($73 million) to develop their own agentic AI solutions through partnerships with Big Tech firms like Microsoft, LG CNS is positioning itself as a homegrown, cost-effective alternative.

The new model adds to LG CNS’s growing portfolio of AI services, which already includes integrations with LG Group’s proprietary LLM, Exaone, as well as global platforms, like OpenAI’s ChatGPT, Meta’s Llama and Alibaba’s Qwen.

Write to Kyung-Ju Kang at qurasoha@hankyung.com

Sookyung Seo edited this article.