Naver on Feb. 20 said it has completed an update to its flagship HyperCLOVA X foundation model and internally released it. The company is now leveraging the model to develop a range of AI services for users, sellers, creators, and business partners.
The updated HyperCLOVA X model is significantly smaller, with about 40% of the parameters of its predecessor, yet delivers more powerful performance. In tests across 19 benchmark categories including Korean, English, coding, and mathematics, it outperformed the previous model in all areas. In the globally recognized MMLU (Massive Multitask Language Understanding) benchmark, it achieved a 79.6% accuracy rate, demonstrating language comprehension on par with similar-sized AI models from global tech giants.
The model’s multi-modality has been improved. It now processes text and image data more effectively than the previous model, enhancing visual question answering and chart comprehension to match top global models. It also added the ability to understand video content.
Operating costs have been reduced by more than 50% in the new HyperCLOVA X model, thanks to its efficient design. Under the On-Service AI strategy announced last year, Naver plans to accelerate AI integration across key services, including search and commerce, leveraging this cost-effective, high-performance model.
Since introducing HyperCLOVA X in August 2023, Naver has continually advanced the model to keep pace with global AI trends. In April last year, Naver introduced HyperCLOVA X DASH, a lightweight version designed for simpler tasks, priced at one-fifth the cost of the original model, helping lower barriers for businesses adopting generative AI. Later in August, the company launched HyperCLOVA X Vision, capable of processing both text and image data simultaneously.
Looking ahead, Naver plans to further enhance the model’s planning and reasoning capabilities, enabling it to perform tasks more comprehensively and systematically. A new version that supports natural voice conversations is set to be released later this year.
Sung Na-ko, head of Hyperscale AI Technology at Naver Cloud, said, “The ability to operate high-performance AI models at low costs has garnered attention, and Naver has consistently developed these technologies to ensure stable AI integration into services used by millions.” He added, “We expect the new HyperCLOVA X model to serve as an engine providing differentiated AI experiences to more users. Going forward, we will continue to enhance core capabilities, such as reasoning and modality expansion, to build AI technologies capable of competing globally.”
To provide users with a more intuitive experience, the upgraded HyperCLOVA X foundation model will be integrated into Naver’s conversational AI service, CLOVA X, in March. Naver also plans to make it available to business clients through Naver Cloud’s hyperscale AI development tool, CLOVA Studio.
