"If corporations that provide products or services with more than 10 million monthly active users (MAU) want to use HyperCLOVA X SEED for commercial purposes, they must request a license from Naver."
"To sell Kanana-2 in an on-device (embedded) form or to sell it to client companies as an on-premises (deployed) solution, a separate license from Kakao is required."
These are the caveats attached as Naver and Kakao released their small language models (SLMs) as open source. Kakao recently open-sourced Kanana-2 and introduced it by saying that it "applies a license that can be used commercially, allowing anyone—including developers, startups, and research institutions—to use the technology without separate restrictions." Naver open-sourced HyperCLOVA X SEED in Apr. last year and said it "can be used free of charge for commercial purposes, which is expected to further vitalize the domestic artificial intelligence (AI) ecosystem." Both companies profess an "open source available for commercial use," but each added its own caveats. Naver singled out competitors, while Kakao singled out business areas such as on-device and locked down commercial use.
◇Most license terms are the same, but they split on "conditions for commercial use"
An analysis by ChosunBiz on the 4th of the license terms for HyperCLOVA X SEED and Kanana-2 posted on the open-source AI platform Hugging Face found that most of the provisions are similar. From the clauses defining how far to consider derivative models, to the obligations when redistributing the model, the attribution of rights to outputs generated by the model, and how to handle disputes, the two documents are essentially written in the same framework.
The two documents diverge in the clause on "additional conditions for commercial use." Naver blocked use by competitors. First, corporations with more than 10 million MAU, or their affiliates, cannot use HyperCLOVA X SEED commercially. Platform corporations such as Kakao, Google, Instagram, Coupang, Baemin, and Karrot fall under this. Second, commercial use is not allowed when distributing products or services that directly compete with, or are substantially similar to, products or services provided by Naver. They must separately request a license from Naver, and whether it is granted is at Naver's discretion.
Kakao blocked commercial use by designating business areas. It restricts cases where ▲the right to access Kanana-2 is sold, ▲Kanana-2 is sold to client companies as an on-premises or system integration (SI) solution, or ▲Kanana-2 is sold in an on-device form. If a case falls under these, a separate commercial license must be obtained from Kakao. Whether it is granted is at Kakao's discretion.
For example, assuming Baemin seeks to implement an "AI review summary" function and apply it to its application (app), Kanana-2 can be used for free, but HyperCLOVA X SEED cannot be used for free due to the "10 million MAU" clause. Conversely, HyperCLOVA X SEED draws a line only for some platform corporations and competitors while leaving the rest open, so it does not explicitly prohibit specific business areas.
◇What Naver protected was the competitive landscape; what Kakao protected was revenue channels
Looking at the caveats, it becomes clear what Naver and Kakao sought to defend even as they released their small language models (SLMs) as open source. Both companies secured the justification of contributing to the developer base and the domestic AI ecosystem by opening their models, yet they protected the cores of their own businesses through license provisions.
Naver reflected in its provisions the need to block situations where its model is used as the engine of a competing service. In addition to dominating domestic search, Naver holds top-tier services in every area: Smart Store and Naver Shopping in commerce, Webtoon/Series/CHZZK in content, and Naver Cloud in corporate infrastructure. Because its fronts of competition are wide, what it is trying to protect is the "competitive landscape" itself.
Kakao also spans commerce, content, and cloud, but these businesses are largely layered onto the single super app KakaoTalk. The power to retain users stems more from KakaoTalk as the national messenger than from the competitiveness of individual services. Thus, rather than competitors building similar services with its model, what Kakao is wary of is "revenue channels" where the model it opened becomes someone else's money-maker. While Kakao's current AI strategy focuses on consumer services such as its tie-up with OpenAI, its provisions are seen as mechanisms to secure room to move into the on-device and on-premises markets, where corporate demand is strong. Kakao says it designed Kanana-2 optimized for on-device environments.
Lee Kyung-jeon, a professor in the Department of Big Data Application at Kyunghee University, analyzed that "the caveats differ because the two models aim for different business models." Lee said, "Naver aims to serve HyperCLOVA X SEED in an interface (API) form, so it blocked competitors that would siphon off traffic," and added, "Kakao aims to optimize Kanana-2 for on-device use, leaving the door open up to the stage of building prototypes for noncommercial use, but adopting a structure where payment is required when moving into full-fledged business."
Wi Jeong-hyeon, a professor in the Department of Business Administration at Chung-Ang University, analyzed that "the differences in the two companies' caveats reflect the gap in their AI capabilities." Wi said, "Naver's provisions imply, 'we are confident in our AI capabilities in the domestic market, so please use it widely except for major competitors,'" adding, "removing only competitors is an aggressive approach." Wi continued, "Kakao's provisions draw the line that 'we will take on businesses that can make money,' which appears to be a much more defensive approach."