HyperCLOVA X Goes Open Source

Naver Cloud recently announced a groundbreaking change in the domestic artificial intelligence ecosystem by fully releasing its HyperCLOVA X lightweight models as commercial free open-source at a tech meetup. Starting on the 24th, the HyperCLOVA X SEED 3B, 1.5B, and 0.5B models will be opened without restrictions to businesses and research institutions, paving the way for the free utilization of AI technology in business and academic research without the burden of GPU resources. This goes beyond the limitations of domestic AI models, which have previously been confined to research use, and holds the potential to drive real industrial innovation. Particularly noteworthy is HyperCLOVA X SEED 3B, a vision-language model (VLM) that understands not just text but also images and videos, boasting advanced visual information processing capabilities such as chart analysis, object recognition, and photo description. It has outperformed similarly sized foreign models in 9 benchmarks related to Korean language, Korean culture, and English visual information understanding, and even demonstrated comparable performance to much larger foreign models with significantly more parameters, showcasing its technological prowess. Kim Yu-won, CEO of Naver Cloud, emphasized that this open-source release is part of Naver's efforts to expand the AI base through cost-efficient specialized models produced via Naver's "On-Service AI Strategy," particularly lightweight models that reflect the high demands of businesses. This signifies a strong commitment to invigorating the domestic AI ecosystem by enabling the use of AI without the burden of GPU resources. At the tech meetup, a future AI roadmap was also presented. The reasoning model, based on the HyperCLOVA X flagship model to be released in the first half of the year, will go beyond accurate answers in mathematics and programming. It will integrate functionalities such as visual and audio information understanding, automatic web search, API calls, and data analysis to showcase complex problem-solving capabilities. For instance, it will be capable of independently planning and executing complex requests like "recommend and book travel destinations and accommodations in Jeju Island." Furthermore, plans for speech models capable of emotional speech synthesis, speech style analysis, and natural two-way conversation were also unveiled, foreshadowing the expansion of multimodality. CEO Kim emphasized that this effort will play a crucial role in building 'Sovereign AI'. He asserted the importance of establishing a robust AI ecosystem capable of producing innovative AI services, as this task requires national capabilities that are difficult to achieve through the efforts of a single company alone. Through this announcement, Naver presented a firm vision to advance HyperCLOVA X, focusing on multimodal models extended to images, videos, and audio, highly accessible lightweight models, and sophisticated reasoning models. This aims to drive the growth of the domestic AI ecosystem and ultimately strengthen the foundation of domestic Sovereign AI.