HyperCLOVA X to be Released as Free Open Source for Commercial Use, Driving the Expansion of Sovereign AI Ecosystem
HyperCLOVA X to be Released as Free Open Source for Commercial Use, Driving the Expansion of Sovereign AI Ecosystem
HyperCLOVA X to be Released as Free Open Source for Commercial Use, Driving the Expansion of Sovereign AI Ecosystem
- In line with its On-Service AI direction, NAVER is developing a series of practical lightweight, multimodal, and reasoning models
- Multimodal open-source models enhance commercial applicability across various services, offering universally applicable without GPU burden
- Sovereign AI is a national imperative and will help strengthen the AI ecosystem
April 23, 2025
NAVER’s generative AI “HyperCLOVA X” will make history on April 24 as the first major domestic generative AI model to be released as a free, open-source tool available for commercial use.
NAVER Cloud (CEO Kim Yuwon) announced the completion of three lightweight HyperCLOVA X SEED models—3B, 1.5B, and 0.5B—which will be made available as open-source. Businesses and research institutions, both local and international, can download these models and adopt them for their commercial or academic needs.
Unlike previous major models, which were limited to research use, the newly released HyperCLOVA X SEED models can now be used for commercial purposes free of charge. This is expected to further invigorate Korea’s AI ecosystem, particularly among small and medium-sized businesses that previously hesitated because of concerns over licensing restrictions or costs.
Multimodal Open-Source Models Enhance Commercial Applicability Across Various Services
In particular, “HyperCLOVA X SEED 3B” stands out as a vision-language model capable of processing not only text but also images and videos. It excels in tasks such as chart analysis, object recognition, and image description. In evaluations across nine benchmarks* designed to assess visual understanding of the Korean language and culture as well as English, HyperCLOVA X SEED 3B achieved higher scores than similarly scaled models from leading US and Chinese tech firms and even matched the accuracy of much larger international models with significantly more parameters.
* Video-MME, NAVER TV CLIP, Video-ChatGPT, Perception Test, ActivityNet-QA, KONet, MMBench-Val, TextVQA-Val, Korean VisIT-Bench
The HyperCLOVA X SEED 3B model is ideal for businesses aiming to develop solutions like ad copy generation from product pages or to design tourism services that integrate photo and video content. Its “ultralightweight” siblings, “HyperCLOVA X SEED 1.5B” and “HyperCLOVA X SEED 0.5B,” are also expected to find broad application thanks to their fast processing speeds and low operational costs.
In Line with Its On-Service AI Direction, NAVER is Developing a Series of Practical Lightweight, Multimodal, and Reasoning Models
NAVER Cloud CEO Kim Yuwon said, “With the acceleration of NAVER’s On-Service AI strategy, we’ve been developing competitive, specialized models that can be seamlessly and cost-effectively integrated into NAVER’s services. Recently, the models most in demand have been lightweight models and reasoning models. Among them, we now have open-sourced lightweight models that businesses can use most universally without the burden of GPU resources.”
NAVER also announced plans to unveil a new reasoning model based on HyperCLOVA X’s flagship version in the first half of the year. This upcoming model will not only improve performance in mathematics and programming—the fundamental strengths of reasoning models—but will also enhance HyperCLOVA X’s capabilities in areas such as visual and audio comprehension, automated web searches, API calls, and data analysis.
For instance, when given a prompt like “Find child-friendly attractions in Seogwipo, Jeju Island, and book a well-rated hotel,” the model autonomously formulates a response plan based on inference. At each stage, it calls search APIs and accommodation booking APIs to complete the task. By linking various types of APIs, such as product purchasing, business automation, and smart home control, and expanding its capabilities in information search and data visualization, the reasoning model can handle increasingly complex instructions. Previously, users had to manually specify which tools an AI should use. Now, with the reasoning model, HyperCLOVA X strengthens its ability to select the appropriate tools for each task on its own.
In addition, NAVER plans to expand HyperCLOVA X’s multimodality from text, image, and video to include voice, with services utilizing the HyperCLOVA X voice model in the pipeline. The HyperCLOVA X voice model incorporates the text model’s knowledge and reasoning capabilities into the voice domain, enabling emotionally rich voice synthesis, voice style analysis, and natural two-way conversations. In the future, NAVER plans to develop an advanced AI conversation model capable of seamlessly switching between text and voice, aiming to create more advanced AI-human voice interactions.
Sovereign AI Is a National Imperative and Will Help Strengthen the AI Ecosystem
The company emphasized that by advancing HyperCLOVA X through multimodal models covering images, videos, and voice, along with low-cost lightweight models and powerful reasoning models, it will drive the expansion of Korea’s AI ecosystem. This will be achieved through applications across NAVER’s large user-base services, enterprise solutions, and open-source initiatives.
CEO Kim Yuwon stated, “Sovereign AI, which is gaining global attention, is a challenge that cannot be achieved by the efforts of a single company alone. It requires comprehensive national capabilities.” He added, “It is essential to build a strong AI ecosystem that can produce innovative AI services closely integrated into daily life, not just focus on securing technology.”
He concluded, “We hope these efforts to expand the AI base will serve as a foundation for further strengthening the resilience of Korea’s sovereign AI ecosystem.” (/End)
