LG’s Exaone AI model matches DeepSeek in cost efficiency: Epoch AI

LG Group headquarters in Yeouido, western Seoul (Newsis)

LG AI Research’s hyperscale artificial intelligence multimodal model has been recognized as notable AI by a leading US research institution, especially receiving high marks for cost efficiency, similar to China’s DeepSeek.

According to a recent report released by Epoch AI, LG’s Exaone 3.5 32B model, unveiled in December, has been included in its "notable AI category."

Epoch AI, composed of scientists and AI experts, evaluates and selects AI models based on performance, usage frequency and other criteria. This marks the first time in two years that a Korean company’s AI model has been recognized in this category.

LG AI Research head Bae Kyung-hoon said at a conference hosted by the Ministry of Science and ICT that the model was developed with an investment of about 7 billion won ($4.8 million). This amount is lower than the estimated $6 million that DeepSeek reportedly spent on developing its V3 model.

LG’s AI research team has optimized the model for practical industrial applications. The initial Exaone 1.0, launched in late 2021, featured 300 billion parameters, but the latest 3.5 model has been scaled down to 32 billion parameters, improving efficiency and reducing costs.

The model is currently deployed in LG Group’s generative AI service, ChatExaone, and its employees use it.

In addition to LG, Korean AI models from Naver and Samsung Electronics, as well as Seoul National University, Korea Advanced Institute of Science and Technology, Yonsei University and Korea University, have been included in Epoch AI’s latest list of notable AI models.