NAVER Builds a ‘New Model’ of Low-Cost, High-Performance HyperCLOVA X: “A New Engine for ‘On-Service AI’”
NAVER Builds a ‘New Model’ of Low-Cost, High-Performance HyperCLOVA X: “A New Engine for ‘On-Service AI’”
NAVER Builds a ‘New Model’ of Low-Cost, High-Performance HyperCLOVA X: “A New Engine for ‘On-Service AI’”
- The latest model is 40% smaller than its predecessor but offers enhanced performance, showcasing language understanding capabilities that rival similar-scale AI models developed by leading tech firms
- "Operational Costs Cut by Over 50%—Enhancing AI Efficiency for Services Used by Tens of Millions, with Ongoing R&D in 'Low-Cost, High-Performance' Models for Stable Deployment"
February 20, 2025
NAVER has completed the update of its HyperCLOVA X flagship model and unveiled it internally on the 20th. Utilizing the newly unveiled HyperCLOVA X foundation model, NAVER marks the beginning of a strategic expansion into AI-powered services designed for users, merchants, content creators, and business partners.
The updated HyperCLOVA X model is designed to be more efficient, featuring 40% fewer parameters than its predecessor while delivering significantly enhanced performance. Benchmark testing across 19 key evaluation metrics—covering Korean, English, coding, and mathematics—confirmed that the new model outperformed its previous iteration in all areas. A key highlight is its performance on “Massive Multitask Language Understanding (MMLU),” a globally recognized AI benchmark, where it achieved an accuracy rate of 79.6%, putting it on par with leading AI models of similar scale from major global tech companies.
The model’s “multimodality” capability has also advanced further. While the previous model could simultaneously process text and image data, including visual question answering and understanding charts and graphs, the updated model has been further optimized to perform at a global top-tier level. Moreover, a new video comprehension feature has been introduced, expanding its processing capabilities beyond images.
Additionally, the model has been designed with an optimized structure, reducing operational costs. According to NAVER, operational costs for the new HyperCLOVA X model have been improved by more than 50% compared to its predecessor. Under its “On-Service AI” strategy announced last year, NAVER has been integrating generative AI technology into key services with large user bases, such as search and commerce. With the new cost-effective, high-performance HyperCLOVA X model, NAVER is expected to accelerate AI adoption even further.
Since unveiling HyperCLOVA X in August 2023, NAVER has continuously upgraded the model in line with global AI technology trends. In April 2024, NAVER launched “HyperCLOVA X DASH,” a lightweight model that operates at one-fifth the cost of the previous model and is optimized for less complex tasks, lowering the barrier for businesses to adopt generative AI. Subsequently, in August, NAVER introduced “HyperCLOVA X Vision,” a model capable of processing both text and images simultaneously.
Furthermore, NAVER is working on advanced planning and reasoning capabilities to enable systematic and comprehensive task execution based on user commands. In the second half of the year, it plans to unveil a new HyperCLOVA X model with natural speech interaction capabilities.
Sung Nako, Executive Director of Hyperscale AI Model at NAVER Cloud, stated, “Recently, the industry has been focusing on technologies that enable high-performance AI models to operate at lower costs. NAVER, which serves tens of millions of users daily, has been dedicated to researching and developing such solutions to ensure AI is seamlessly integrated into its services.” He further stated, “We expect the new HyperCLOVA X model to power an enhanced AI service experience for a broader audience. Going forward, we will continue strengthening our core AI models by advancing inference capabilities and expanding multimodal functions, building AI technology that can compete on a global scale.”
Meanwhile, NAVER plans to integrate the upgraded HyperCLOVA X foundation model into its CLOVA X conversational AI service in March, enabling users to experience the enhanced model firsthand. Additionally, NAVER plans to release the model through the hyperscale AI development tool “CLOVA Studio” on the NAVER Cloud Platform, making it available to business customers as well. </End>
