Korea’s ambition to build sovereign AI capabilities is taking even further in tangible form. KONAN Technology, a Seoul-based AI firm long known for its text and vision software, has emerged as the nation’s first developer of a homegrown large language model (LLM). Its recent achievements position Korea not only to compete with China’s DeepSeek and Western models but also to redefine how nations approach AI independence.

KONAN Unveils Korea’s First Homegrown LLM, Beating DeepSeek in Efficiency

KONAN Technology has developed Korea’s first independently built large language model, “KONAN LLM,” advancing the country’s pursuit of sovereign AI capabilities.

Founded in 1999, the company began focusing on generative AI in 2022 and became the first in Korea to deploy NVIDIA’s H100 GPUs for model training. By 2023, KONAN released its own 13.1-billion-parameter LLM — a milestone establishing domestic capacity to build and scale foundation models without external dependencies.

In 2025, KONAN introduced the ENT-11 model, Korea’s first LLM integrating general-purpose and reasoning functions in a single architecture. Unlike competitors such as Gemini that require users to switch between separate modes, ENT-11 operates both simultaneously — cutting GPU usage by nearly half while maintaining advanced reasoning accuracy.

Despite its smaller scale of 32 billion parameters — over 20 times smaller than China’s DeepSeek R1 (671B) — ENT-11 delivered comparable conversational and reasoning performance and even outperformed DeepSeek in coding accuracy by 4.75 percentage points.

KONAN Technology: From Text AI to Full-Stack AI Sovereignty

KONAN began as a text and vision AI company before transitioning into full-stack AI research. Executive Director Choi Jung-joo of KONAN’s Vision AI Research Institute described the company’s evolution as a reflection of Korea’s AI policy direction,

“KONAN is a full-stack AI company capable of managing every step — from data curation to model deployment. KONAN LLM was the first large model fully developed domestically, marking an important step toward sovereign AI.”

The company has also launched KONAN RAG-X, a retrieval-augmented generation solution that links enterprise internal data with real-time external sources for more accurate, up-to-date responses.

The solution earned a Grade 1 Good Software (GS) certification and was designated a priority procurement product by Korea’s Public Procurement Service — signaling government trust in its enterprise applications.

Efficiency, Security, and Practical AI

Executive Director Choi Jung-joo further emphasized KONAN’s focus on practical, secure AI systems that meet national and enterprise needs.

“Traditional AI models rely on past data, but companies operate with constantly changing information. KONAN RAG-X addresses this by combining precise vector search and long-standing retrieval technologies to achieve both accuracy and data protection.”

KONAN’s model has already seen adoption across key sectors, including energy and defense. The firm recently secured a project with the Korean military to develop a domain-specific LLM that supports intelligence-based decision-making in secure environments.

“In defense projects, data security often prevents external developers from accessing sensitive information. Our proposal included not only a foundation model but also a software package that enables fine-tuning within the client’s environment — effectively teaching how to fish, not just providing the fish.”

A Strategic Milestone for Korea’s AI Competitiveness

KONAN’s breakthrough arrives at a critical juncture in Korea’s AI strategy. As the government promotes sovereign AI development to reduce dependence on U.S. and Chinese models, KONAN Technology demonstrates that domestic firms can achieve global benchmarks in efficiency and performance.

The company’s success also aligns with national efforts under the Ministry of Science and ICT and the Ministry of SMEs and Startups (MSS) to accelerate the commercialization of Korean AI across defense, public administration, and enterprise innovation sectors.

By developing compact yet high-performing models like ENT-11, Korea positions itself competitively within the global AI ecosystem — not by scale alone, but through resource optimization, compliance readiness, and enterprise integration.

Choi noted that KONAN plans to expand its model architecture to large action models (LAMs), supporting domain-specific multi-agent collaboration and automation across industries.

KONAN’s Global Push: Expanding Korea’s AI Presence in ASEAN

Looking beyond domestic markets, KONAN is targeting Southeast Asia — a region with growing AI demand but limited resources. Choi Jung-joo explained,

“The ASEAN market has strong interest in AI but faces challenges in data and capital. We are partnering with local universities and research institutes to develop foundation models suited for the region’s needs.”

This approach reflects Korea’s broader strategy of exporting digital innovation through AI partnerships, data collaboration, and regional R&D integration — positioning its startups as technology providers rather than just adopters.

KONAN Techbology: A Defining Moment for Korea’s AI Sovereignty

KONAN Technology’s ascent signals a pivotal transformation in Korea’s innovation landscape — from an AI consumer to a technology creator.

As countries race to build their own sovereign AI infrastructures, KONAN’s model demonstrates that Korea’s approach — focusing on efficiency, reliability, and sector-specific utility — may define a new paradigm of “smart sovereignty” in AI.

If sustained through global partnerships and policy support, KONAN’s progress could mark the beginning of Korea’s emergence as an independent force in the global AI race.

– Stay Ahead in Korea’s Startup Scene –
Get real-time insights, funding updates, and policy shifts shaping Korea’s innovation ecosystem.
➡️ Follow KoreaTechDesk on LinkedIn, X (Twitter), Threads, Bluesky, Telegram, Facebook, and WhatsApp Channel.