HyperCLOVA X Surpasses 300,000 Downloads Just One Month After Open-Source Launch, Expanding Global Influence
HyperCLOVA X Surpasses 300,000 Downloads Just One Month After Open-Source Launch, Expanding Global Influence
HyperCLOVA X Surpasses 300,000 Downloads Just One Month After Open-Source Launch, Expanding Global Influence
- Monthly downloads rank in the top 0.03% among open-source models worldwide; 30 derivative models are also released
- Praised by developer communities for its on-device usability and Korean language performance
- Recognized globally for its excellent understanding of the Korean language and culture, contributing to greater AI diversity
May 26, 2025
NAVER Cloud (CEO Kim Yu-won) announced that the three open-source AI models of “HyperCLOVA X SEED,” released on April 24, have surpassed 300,000 downloads in just over a month. This milestone places the models in the top 0.03% of all models registered on “Hugging Face,” a global AI model-sharing platform, highlighting the expanding global influence of HyperCLOVA X.
The visual-language model “HyperCLOVA X SEED 3B,” capable of processing text, image, and video information together, recorded approximately 267,000 downloads, receiving the most enthusiastic response among the three models. In a comparison of average scores across nine benchmark tests designed to evaluate visual information comprehension in Korean and English, the HyperCLOVA X SEED 3B model outperformed similarly sized models from major US and Chinese tech companies. It demonstrated accuracy rates comparable to much larger overseas models with significantly more parameters, thereby proving its performance competitiveness.
* VideoMME, NAVER-TV-CLIP, VideoChatGPT, Perception Test, ActivityNet-QA, KoNet, MMBench-Val, TextVQA-Val, Korean VisIT-Bench
The text models “HyperCLOVA X SEED 1.5B” and “HyperCLOVA X SEED 0.5B” have also recorded approximately 35,000 downloads. Notably, 29 derivative models have been created based on the 2 models, leading to various improvements and experiments. Positive user feedback has been shared in developer communities, including comments such as: “Despite its small size, the model demonstrates balanced language performance and excellent stability”; “even with just 1GB of RAM, a smart Korean language model can run on-device”; and “with only supervised fine-tuning (SFT) and low-rank adaptation (LoRA) training, the model responds very well. It is fast even in a central processing unit (CPU) environment, making it ideal for building internal chatbots with ease.”
Sung Nako, Head of Hyperscale AI Technology at NAVER Cloud, stated, “It is meaningful that Team NAVER’s generative AI model, developed entirely from scratch, is gaining attention in the global open-source community.” He added, “We hope that HyperCLOVA X can contribute to expanding diversity within the AI ecosystem, where models that understand various languages and cultures, including Korean, can coexist. Ultimately, it will serve as a foundation for more innovative technologies and services.”
Meanwhile, the company plans to unveil its in-development HyperCLOVA X reasoning model through its conversational AI service “CLOVA X” in June. In addition, HyperCLOVA X’s multimodality, which has already expanded from text to images and videos, will be further extended to voice. Services utilizing the HyperCLOVA X voice model are also planned to be introduced sequentially in the future. (/End)
