LG unveils new inference AI model, EXAONE Deep

SEOUL, March 18 (Yonhap) -- LG AI Research, the artificial intelligence (AI) lab under South Korea's LG Group, said Tuesday it has unveiled its new inference AI model, EXAONE Deep, as part of its push for agentic AI.

EXAONE Deep was unveiled at the GTC 2025 hosted by AI chip giant Nvidia Corp. from Monday through Friday in California and released as an open-source model, according to LG AI Research.

Agentic AI uses sophisticated reasoning and iterative planning to autonomously solve complex, multistep problems. To bring agentic AI into practice, more advanced AI technologies, such as inference AI, are needed.

The company said EXAONE Deep is currently the only South Korean model capable of competing with other global inference AI models. Currently, a handful of major global tech firms with foundation AI models, such as OpenAI's GPT series, have inference AI.

Along with the main EXAONE Deep-32B, LG AI Research also released its lightweight EXAONE Deep-7.8B and on-device EXAONE Deep-2.4B models as open-source models.

In a technical report evaluation, the main model's performance received top scores in mathematics, science and coding, compared with global open-source AI models from the United States, China and other countries, according to the company.

LG AI Research said it plans to lead the on-device AI market by upgrading its AI models through collaboration with LG Group affiliates, including LG Electronics Inc. and LG Uplus Corp.

brk@yna.co.kr
(END)