LG AI Research Releases EXAONE 2.0 750B Model on Hugging Face

LG AI Research has officially made its flaghip 750-billion-parameter foundation model, EXAONE 2.0, publicly available on Hugging Face. The open-weights release marks one of the largest foundation models ever made accessible for commercial and academic research, providing high-speed inference across complex multimodal tasks.

Pre-trained on over 3 trillion tokens spanning technical literature, patent databases, and high-quality bilingual corpora, EXAONE 2.0 excels in dual English-Korean reasoning. The model architecture incorporates advanced mixture-of-experts (MoE) routing to maintain high throughput on enterprise hardware.

By granting open access to EXAONE 2.0, LG aims to foster a global open-source AI ecosystem, empowering researchers and developers to fine-tune specialized domain models without prohibitive upfront pre-training investments.