From tech giants to startups, South Korean gamers are growing giant language fashions tailor-made to their very own language and tradition, able to compete with international heavyweights like OpenAI and Google.
Last month, the nation launched its most bold sovereign AI initiative thus far, pledging ₩530 billion, (about $390 million), to 5 native firms constructing large-scale foundational fashions.
The transfer underscores Seoul’s want to chop reliance on international AI applied sciences, hoping to strengthen nationwide safety and maintain a tighter management over knowledge within the AI period.
The organizations picked by the Ministry of Science and ICT to compete have been LG AI Research, SK Telecom, Naver Cloud, NC AI, and the startup Upstage.
Each six months, the federal government will assessment the primary cohort’s progress, reduce underperformers, and proceed funding the frontrunners till simply two stay to steer the nation’s sovereign AI drive.
Every participant is bringing a distinct benefit to South Korea’s AI race. TechCrunch spoke with a number of of the chosen firms about how they plan to tackle OpenAI, Google, Anthropic and the remaining on their house turf. NC AI declined to remark.
LG AI Analysis: Exaone
LG AI Analysis, the R&D unit of South Korean large LG Group, provides Exaone 4.0, a hybrid reasoning AI mannequin. The most recent model blends broad language processing with the superior reasoning options first launched within the firm’s earlier Exaone Deep mannequin.
Techcrunch occasion
San Francisco
|
October 27-29, 2025
Exaone 4.0 (32B) already scores fairly nicely in opposition to opponents on Artificial Analysis’s Intelligence Index benchmark (as does Upstage’s Solar Pro2). But it surely plans to enhance and transfer up the ranks by its deep entry to real-world {industry} knowledge starting from biotech to superior supplies and manufacturing.
It’s coupling that knowledge with a give attention to refining the information earlier than feeding to the fashions to coach. As an alternative of chasing sheer scale, LG needs to make the whole course of extra clever, so its AI can ship actual, sensible worth that goes past what general-purpose fashions can provide. “That is our elementary strategy,” co-head Honglak Lee instructed TechCrunch.
LG is bettering its fashions through acquainted techniques: providing them by APIs, then utilizing the real-world knowledge generated by customers of these providers to coach the mannequin to enhance.
“As LG’s fashions enhance, our companions can ship higher providers, which in flip generate better financial worth and even richer knowledge,” he stated.
Nonetheless, as an alternative of chasing large GPU clusters, LG AI Analysis is specializing in effectivity, getting probably the most out of each chip, and creating industry-specific fashions, he talked about. The aim isn’t to outspend the worldwide giants however to outsmart them with high-performing, but extra environment friendly, AI.
South Korea’s telco large SK Telecom (SKT) launched its personal AI agent A. (pronounced A-dot) service means again in late 2023 and simply rolled out its new giant language mannequin, A.X, this July.
Constructed on high of the Chinese language open supply mannequin from Alibaba Cloud, Qwen 2.5, A.X 4.0 is available in two fashions, a hefty 72-billion-parameter model and a lighter 7B model.
SK says that A.X 4.0 processes Korean inputs about 33% extra effectively than GPT-4o did, underscoring its native language edge. (OpenAI’s GPT 5.0 comparability knowledge isn’t obtainable.) SKT additionally open sourced its A.X 3.1 models earlier this summer time. In the meantime, the A. service provides options like AI name summaries and auto-generated notes. As of August 2025, it’s already pulled in about 10 million subscribers.
SK’s edge is its versatility, as a result of it has entry to data from its telecom community starting from navigation to taxi-hailing.
“SK Telecom’s function is to behave as a bridge between cutting-edge mannequin analysis and real-world impression. With our telecom infrastructure, in depth person base and confirmed service like A., we deliver AI straight into on a regular basis life, whether or not in customer support, mobility, or manufacturing,” Taeyoon Kim, head of the muse mannequin workplace at SK Telecom, instructed TechCrunch.
SK Telecom can also be investing in AI infrastructure, utilizing GPUaaS, South Korea’s largest GPU-based service, and constructing a brand new hyperscale AI knowledge heart with AWS. No matter it lacks, it’s partnering to acquire.
“We’re constructing a full-stack ecosystem with Korean AI chipmaker Rebellions, securing trusted knowledge partnerships by work with the federal government and universities, and fostering a worldwide analysis community,” stated Kim. “That features initiatives like our collaboration with MIT (MGAIC), which applies basis fashions to superior manufacturing and battery and semiconductor innovation.”
Naver Cloud: HyperCLOVA X
Naver Cloud, the cloud providers arm of South Korea’s main web firm, launched its giant language mannequin, HyperClova, in 2021. Two years later, it unveiled an upgraded model, HyperCLOVA X, together with new merchandise powered by the expertise: CLOVA X, an AI chatbot, and Cue, a generative AI-driven search engine positioned as a rival to Microsoft’s CoPilot-enhanced Bing and Google’s AI Overview. It additionally unveiled this 12 months its multimodal reasoning AI mannequin, HyperCLOVE X Think.
Naver Cloud believes the true energy of LLMs is to function “connectors” linking legacy techniques and siloed providers to enhance usefulness, in accordance with a Naver spokesperson.
Naver stands out as Korea’s solely firm — and one of many few on this planet — that may genuinely declare to have an “AI full stack.” It constructed its HyperCLOVA X mannequin from scratch and runs the large knowledge facilities, cloud providers, AI platforms, functions, and shopper providers that deliver the expertise to life, the spokesperson defined.
Just like Google — however tuned for South Korea — Naver is embedding its AI into core providers like search, procuring, maps, and finance. Its benefit is real-world knowledge. It’s AI Procuring Information, as an example, provides suggestions based mostly on what folks truly need to purchase. Different providers embody CLOVA Studio, which lets companies construct customized generative AI, and CLOVA Carecall, an AI-powered check-in service geared for seniors residing alone.
The Naver spokesperson says besting international AI giants like OpenAI and Google hinges on two issues: perfecting its “recipe” for fashions and securing the capital to scale them. Even so, reasonably than chasing measurement, the corporate emphasizes sophistication, arguing its AI is already globally aggressive at comparable scales.
Upstage’s Photo voltaic Professional 2
Upstage is the one startup competing within the venture. Its Photo voltaic Professional 2 mannequin, launched final July, was the primary Korean mannequin acknowledged as a frontier mannequin by Artificial Analysis, placing it within the ring with OpenAI, Google, Meta, and Anthropic, in accordance with Quickly-il Kwon, government vice chairman at Upstage.
Whereas most frontier fashions have 100 billion to 200 billion parameters, Photo voltaic Professional 2 — with simply 31 billion — performs higher for South Koreans and is less expensive, Kwon instructed TechCrunch.
“Photo voltaic Professional 2 has outperformed international fashions on main Korean benchmarks. With this venture, Upstage goals to realize a Korean language efficiency of 105% of the worldwide customary,” Kwon stated.
Upstage goals to distinguish itself by specializing in actual enterprise impression, not simply benchmarks, he stated. So it’s growing specialised fashions for industries like finance, legislation, and drugs, whereas pushing to construct a Korean AI ecosystem led by “AI-native” startups.