Naver has unveiled an upgraded version of its large language model (LLM), HyperClova X, featuring a lightweight design with a 40% reduction in parameters compared to its predecessor. This strategic move aims to compete with the rise of cost-effective AI models from Chinese firms like DeepSeek.

Despite the significant reduction in parameters, Naver reports enhanced performance in the new model. On the MMLU benchmark, a comprehensive evaluation of AI intelligence across 57 subjects, HyperClova X achieved an impressive accuracy rate of 79.6%, surpassing the previous model's 68.0%.

This achievement positions Naver's AI competitively with global tech giants. Estimated to be built with 30 to 40 billion parameters, it outperforms similar-sized models like Alibaba's Qwen 2.5 (75.2%) and Apple's lightweight AI (63.7%). While trailing Mistral AI's Mistral Small 3 (81%), Naver secures a leading position among lightweight AI models.

The upgraded HyperClova X also boasts enhanced multi-modality capabilities, enabling seamless processing of diverse visual data formats, including videos, charts, and diagrams, in addition to text and images. Naver claims a 50% reduction in operating costs for the new model compared to its predecessor.

This upgrade aligns with Naver's broader strategy to integrate AI across its entire suite of services, marking a significant advancement in lightweight AI technology.