South Korea

AI models, vendors, and hardware buying realities from South Korea. Curated by myaihardware.com.

16
Models
9
Vendors
2
Languages

Why South Korea matters for AI builders

South Korea is one of the few places on earth where you can buy high-bandwidth memory (HBM) and advanced foundry capacity for AI chips outside of TSMC's Taiwan, a hardware-availability reality that directly de-risks your supply chain. Its AI builders overwhelmingly deploy open-weight models like Llama and Mistral, not proprietary APIs, making the market a strong signal for inference-heavy hardware demand. The underrated angle is that Korea's dense data center buildout is driven by domestic hyperscalers (Naver, Kakao) not just Samsung/LG, creating a unique channel for edge and HPC hardware that ignores the US-centric narrative.

Market context

South Korea's sovereign-AI drive is propelled by Naver, LG, and Samsung building domestic foundation models alongside government chip subsidies that aim to reduce reliance on foreign hardware. The evaluation landscape has rapidly matured around native benchmarks like KMMLU, HAERAE, and LogicKor, which capture linguistic and cultural nuance far better than translated English tests. Underlying these efforts, traditional byte-pair encoding stumbles on Hangul's dense syllabic blocks within the CJK range, spurring bespoke Korean-first tokenizers that optimize morphology by treating jamo sequences rather than raw bytes.

Models (16 of 16)

Filter by vendor, license, or parameter range.

HyperCLOVA X HCX-003

Naver · South Korea

Restricted
Params
204B
Context
65,536
Min VRAM
200 GB
GGUF
No

Exceptional Korean text generation with cultural nuance.

Region slug: korea

HCX SEED

Naver · South Korea

Restricted
Params
13B
Context
32,768
Min VRAM
10 GB
GGUF
No

Efficient and fast for real-time applications.

Region slug: korea

EXAONE 3.0 7.8B

LG AI Research · South Korea

Open
Params
7.8B
Context
4,096
Min VRAM
8 GB
GGUF
Yes

Excellent bilingual performance with efficient size.

Region slug: korea

EXAONE 3.5 2.4B

LG AI Research · South Korea

Open
Params
2.4B
Context
2,048
Min VRAM
3 GB
GGUF
Yes

Very lightweight and fast, good for on-device.

Region slug: korea

EXAONE 3.5 7.8B

LG AI Research · South Korea

Open
Params
7.8B
Context
4,096
Min VRAM
8 GB
GGUF
Yes

Strong reasoning improvements over 3.0.

Region slug: korea

EXAONE 3.5 32B

LG AI Research · South Korea

Open
Params
32B
Context
8,192
Min VRAM
16 GB
GGUF
Yes

Top-tier performance among open Korean-English models.

Region slug: korea

EXAONE Deep

LG AI Research · South Korea

Restricted
Params
32B
Context
32,768
Min VRAM
16 GB
GGUF
Yes

Efficient inference via MoE, excels at reasoning.

Region slug: korea

SOLAR 10.7B

Upstage · South Korea

Open
Params
10.7B
Context
4,096
Min VRAM
8 GB
GGUF
Yes

Packs 13B-level performance in a smaller size.

Region slug: korea

SOLAR Mini

Upstage · South Korea

Open
Params
1.7B
Context
2,048
Min VRAM
2 GB
GGUF
Yes

Extremely fast and low-resource.

Region slug: korea

SOLAR Pro

Upstage · South Korea

Restricted
Params
22B
Context
4,096
Min VRAM
16 GB
GGUF
Yes

Best performance among SOLAR models, good Korean.

Region slug: korea

Mi:dm 2.0

KT Corporation · South Korea

Restricted
Params
40B
Context
4,096
Min VRAM
40 GB
GGUF
No

Deep Korean linguistic knowledge from KT's data.

Region slug: korea

A.X 4.0

SK Telecom · South Korea

Restricted
Params
13B
Context
8,192
Min VRAM
12 GB
GGUF
No

Optimized for Korean conversational AI.

Region slug: korea

KoGPT

Kakao · South Korea

Restricted
Params
30B
Context
4,096
Min VRAM
30 GB
GGUF
No

Strong Korean generation from Kakao's data ecosystem.

Region slug: korea

Polyglot-Ko 12.8B

EleutherAI · South Korea

Open
Params
12.8B
Context
2,048
Min VRAM
8 GB
GGUF
Yes

Open and reproducible Korean pretrained model.

Region slug: korea

VARCO

NCSoft · South Korea

Restricted
Params
13B
Context
2,048
Min VRAM
12 GB
GGUF
No

Tailored for game narratives and interactive fiction.

Region slug: korea

42dot LLM

42dot · South Korea

Restricted
Params
7B
Context
4,096
Min VRAM
6 GB
GGUF
No

Optimized for automotive and navigation tasks.

Region slug: korea

Notable vendors

NaverLG AI ResearchUpstageKTSK TelecomKakaoNCSoft42dotEleutherAI

Run benchmarks

llama.cpp, MLPerf, and inference leaderboards.

VRAM calculator

Will this model fit on your GPU?

Hardware guides

Local LLM setup, homelab, mini-PC builds.

Methodology

How we benchmark, verify ASINs, and source data.

Beginner's guide

New to local AI? Start here.

Sourcing & commerce disclosure

Model metadata on this page is compiled from each vendor's public HuggingFace repo, model card, or research paper at time of indexing. Performance figures (KMMLU, MMLU, HAERAE, AraBench, IndicGEM, etc.) link directly to source benchmarks where available; missing or partial values are shown as ", ".

Outbound hardware links (GPUs, mini-PCs, accelerators) are Amazon Associate links using store ID fredoline-20. As an Amazon Associate we earn from qualifying purchases. Prices and availability shown elsewhere on the site are snapshots, not live quotes — verify on Amazon before purchasing. See About / disclosures for the full policy.