HyperCLOVA X

HyperCLOVA X is Naver's flagship hyperscale language model, unveiled August 2023 as South Korea's answer to GPT-4 and one of the first major foundation models trained Korean-first rather than English-first. Naver reports roughly 204B parameters, trained on a balanced mix of Korean, English, and code with SFT and RLHF alignment. It powers Naver's CLOVA X chatbot and the CUE generative search service, and is sold to enterprises via Naver Cloud.

Naver has extended the family with HyperCLOVA X Vision (image understanding), HyperCLOVA X THINK (a reasoning variant with 128K context, competitive on Korean benchmarks like KMMLU and CSAT), and lightweight SEED models including an open 3B vision-language model. In 2026 Naver joined a Korean government consortium building a 700B-parameter security AI on HyperCLOVA X + LG Exaone.

Honest limits: the flagship model is proprietary — no open weights, API/Cloud-only access outside Korea-centric channels. English reasoning trails top US/Chinese frontier models, and pricing/availability outside Naver Cloud is opaque.