Naver Cloud Launches Commercial Open-Source Hyperscale LLM

Naver Cloud has set an innovative milestone in the AI market. It has fully released 'HyperCLOVA X SEED 14B Think', a lightweight inference model developed with its proprietary original technology, as a free open-source available for commercial use. This AI, born from pure domestic technology and not merely a modification of an overseas open-source model, perfectly combines advanced inference capabilities with unique lightweight technology. It has expanded the inference model, which is central to AI agent services, beyond research applications into the business domain. This model, a lightweight version of HyperCLOVA X THINK, has dramatically reduced training costs through sophisticated methods such as efficiently pruning low-importance parameters and transferring knowledge from the original model to a smaller one. What's astonishing is that despite being a 14-billion-parameter model, its training cost is only 1/100th that of global open-source models of the same size. It even achieved this performance with less GPU training time than a 500-million-parameter global model. It is overwhelming not only in cost-efficiency but also in performance. It has proven its excellence by achieving average scores similar to or even surpassing those of overseas models of the same size (14 billion parameters) or larger (32 billion parameters) in evaluations related to Korean language, Korean culture, coding, and mathematics. Naver Cloud expressed a strong aspiration, stating, "HyperCLOVA X, built with our own technology, will lead the growth of the Korean AI ecosystem." Just as the three lightweight HyperCLOVA X models released last April surpassed 1 million cumulative downloads, contributing to active derivative model creation and the spread of on-device AI services, this release is also expected to stand at the forefront of global AI innovation.