LG unveils Korea’s 1st hybrid AI model EXAONE 4.0

LG AI Research, the artificial intelligence (AI) lab under LG Group, introduced Korea’s first hybrid AI model, EXAONE 4.0, Tuesday. The hybrid model combines a large language model (LLM), which excels at understanding and generating human language and providing fast, knowledge-based answers, with a reasoning AI model that can form and test hypotheses independently to solve problems. Currently, only a few companies have released hybrid AI models, such as Anthropic in the United States and Alibaba in China. The lab released EXAONE 4.0 as an open-weight model on a global open-source AI platform, Hugging Face, for research and educational purposes. Open-weight refers to models that have publicly accessible internal variables learned during training, allowing others to modify or redistribute the model. According to the lab, its latest model showed successful performance in various benchmark tests, scoring 92.3 on MMLU-Redux and 81.8 on MMLU-Pro, which measure knowledge and problem-solving. It scored 75.4 in the science-focused GPQA-Diamond test, and 85.3 in the math problem-solving AIME 202Weiter zum vollständigen Artikel bei Korea Times Weiter zum vollständigen Artikel bei Korea Times