Naver Corp.’s generative artificial intelligence (AI), HyperCLOVA X, will become the first among major domestic generative AI models to be released as a free, open-source model available for commercial use. The company also plans to release a separate inference model in the first half of 2025.

At a press conference held on Wednesday at Naver Square in Gangnam, southern Seoul, Naver Cloud Chief Executive Officer Kim Yu-won announced the completion and open-source release of three lightweight versions of HyperCLOVA X: HyperCLOVA X SEED 3B, HyperCLOVA X SEED 1.5B, and HyperCLOVA X SEED 0.5B.

The models will be introduced on Thursday via Hugging Face, a global open-source AI platform and community. Companies and research institutions, both domestic and international, can download and customize these models to suit their business or academic research needs.

“While enhancing our flagship model, we have also slimmed it down to offer a variety of smaller models,” Seong Nak-ho, head of hyperscale AI technology at Naver Cloud, said.

Among the three newly released models, the mid-sized HyperCLOVA X SEED 3B is a visual language model capable of processing not only text but images and video as well. It can understand charts, recognize objects, and describe photos. It can also be used in solutions that analyze product pages, for example, to suggest ad copy or guide users via Korean travel destinations via photo and video content.

Unlike most previous domestic models that were limited to research use, the open-source HyperCLOVA X SEED can be used freely for commercial purposes. This move is expected to invigorate the domestic AI ecosystem, particularly among small and medium-sized businesses that were hesitant to adopt generative AI due to cost and licensing issues.

“As Naver’s On-Service AI strategy accelerates, we are developing competitive, specialized models that can be applied stably and cost-effectively to our services,” Kim said. “Lightweight and inference models are among the models that have recently been in high demand.”

Kim noted that the company has open-sourced the lightweight models that companies can use freely and broadly. Naver Cloud plans to unveil the inference version of HyperCLOVA X, which is currently under development, within the first half of 2025.

This model goes beyond offering more accurate responses in match and programming - which are inference models’ traditional strengths - to include enhanced capabilities in visual and speech understanding, automated web searches, and API calls.

For his part, Kim also emphasized AI sovereignty.

“As we face a fierce U.S.-China tech rivalry and rising uncertainty, it is critical to have the capability to develop and operate AI independently,” Kim said. “AI sovereignty is not a choice but a necessity, which is why national-level support is vital.”

[ⓒ 매일경제 & mk.co.kr, 무단 전재, 재배포 및 AI학습 이용 금지]