South Korea
Open models from South Korea
We track 35 open-weight models from 3 labs in South Korea. They account for 8 of 205 tracked releases (4%) from Dec 2025 to Nov 2026. The newest is K-EXAONE-2.0-750B-A37B-DSpark, listed on 30 Jul 2026.
Explore the models
Latest models
Newest listed firstCheapest tracked APIs
No hosted API prices recorded. The lab pages link to the available weights.
How the work developed
LG AI Research opened an EXAONE instruction-tuned checkpoint for English and Korean. Its model card describes the bilingual model and links the team’s research report and custom licence. Upstage’s SOLAR work used depth up-scaling: extending a model’s layers and continuing pretraining from Mistral weights. Upstage distinguished the Apache-licensed base model from its non-commercial instruction-tuned release.
NAVER added downloadable HyperCLOVA X SEED checkpoints focused on Korean language and culture. The SEED text-model card describes pretraining, supervised fine-tuning and reinforcement learning, with a separate HyperCLOVA X SEED licence. These Korean labs publish different model families with different permissions; open weights do not make their licence conditions interchangeable.
Sources and counting method
Counts cover selected repositories, including model variants. Dates come from Hugging Face repository creation dates and may differ from public announcements. Location follows the lab record, not the nationality of individual contributors.