LG AI Research

Artificial Intelligence · Large Language Models · South Korea Tech

Launch

LG unveils K-EXAONE 2.0, Korea's largest AI model with 750B parameters

Releasing it under a fully permissive license lets any company build commercial products on it without paying royalties.

• LG AI Research released K-EXAONE 2.0 on the open-source platform Hugging Face on July 31, 2026, calling it the largest AI foundation model built in Korea to date.
• LG AI Research is the corporate institute driving LG Group's AI ambitions, and the model is the second release under the Ministry of Science and ICT's Independent AI Foundation Model Project to build homegrown, sovereign AI.
• K-EXAONE 2.0 has 750 billion parameters, more than triple the 236 billion of the project's first model, and carries an Apache 2.0 license that permits unrestricted commercial use with no fees.
• The model averaged 70.1 across 24 benchmarks, up over 10% from the prior model's 63.3, with coding and agentic-coding performance jumping about 30%.
• In long-context comprehension tests it scored 94.4 and 89.6 (English and Korean), beating China's Zhipu GLM-5.1 at 71.5 and 83.6, and it topped GLM-5.1 and Alibaba's Qwen3.5 on an agentic tool-use benchmark.
• Language support expanded from six to ten, adding French, Italian, Portuguese and Polish, positioning it as the most multilingual sovereign AI model built in Korea.
• By open-sourcing a state-backed, benchmark-competitive model under a permissive license, LG is turning a national sovereign-AI initiative into shared infrastructure that Korean firms can build on without licensing costs, a direct counter to China's open-weight model push.