LG unveils hybrid AI

LG Group has launched a new iteration of its flagship AI model, which it claims is South Korea's first hybrid-reasoning AI system. This positions LG as a key player in the intensifying global AI race. The model combines different AI reasoning methods to enhance its performance and problem-solving capabilities. LG is not releasing the model for public use, instead keeping it for internal product development.

The new AI model, named Exaone Deep, includes the main model Exaone Deep-32B, which features 32 billion parameters. While DeepSeek's R1 has 671 billion parameters, Exaone Deep-32B has only about 5% of that. Despite having significantly fewer parameters, performance tests show that LG's model is on par with DeepSeek-R1. In benchmark comparisons, Exaone Deep-32B excelled in mathematics, scoring 90 in the 2024 U.S. Mathematical Olympiad, outperforming DeepSeek-R1 (86.7) and Alibaba's QwQ-32B (86.7). In South Korea's 2025 CSAT math section, it achieved 94.5, the highest among competing models. It also performed well on doctoral-level science problems, scoring 66.1, ahead of Alibaba's QwQ-32B (63.3).

LG AI Research also unveiled its upgraded pathology AI model, EXAONE Path 2.0. Compared with the previous 1.0 version, EXAONE Path 2.0 delivers enhanced accuracy in analysing and predicting genetic mutations, gene expression patterns and subtle structural changes in human cells and tissues, using pathological tissue images. The model has been trained on over 10,000 datasets where whole slide images are paired with multiomics information, enabling it to predict gene activity solely through image analysis without the need for expensive genomic testing.