Epoch AI has recognized Exaone Deep as one of the notable AI models, further solidifying LG's presence in the AI landscape.

This model is expected to excel in specialized evaluations across mathematics, science, and coding, making it particularly beneficial for businesses and educational institutions.

Introduced at the Nvidia GTC conference in Silicon Valley, Exaone Deep boasts impressive performance with only 32 billion parameters, significantly fewer than competitors like DeepSeek R1, which operates on 671 billion parameters.

The model's efficiency is noteworthy, requiring just one Nvidia H100 GPU to achieve comparable performance to DeepSeek, which needs 16 GPUs.

Exaone Deep achieved a remarkable score of 94.5 on the 2025 Korean College Scholastic Ability Test, outperforming DeepSeek R1 and OpenAI's o1-mini.

In addition, it scored 95.7 on the MATH-500 index and 66.1 on the GPQA Diamond test, demonstrating its competitive edge in advanced reasoning.

The model also excelled in the Massive Multitask Language Understanding benchmark, scoring 83 and outperforming OpenAI's o1-mini.

LG has introduced a lightweight version, Exaone Deep-7.8B, which retains 95% of the performance of the larger model while being only 24% of its size.

All Exaone Deep models have been released as open-source, promoting accessibility for researchers and developers.

LG AI Research has unveiled Exaone Deep, Korea's first reasoning AI model, designed to compete with industry leaders like OpenAI and DeepSeek.

Looking ahead, the LG AI Research Institute plans to collaborate with LG Electronics and LG Uplus to develop on-device AI services based on Exaone Deep.

CEO Koo Kwang-mo emphasized the company's commitment to integrating advanced technologies like AI into daily life to enhance convenience and create new value.