KaKao

- Participation in an open source AI technology session to share its in-house model strategy and open source roadmap

- Release of a wide range of AI models as open source since February, surpassing 400,000 cumulative downloads

- Continued efforts to enhance transparency and accessibility in AI technology, contributing to the growth of Korea’s AI ecosystem

 

[November 18, 2025] Kakao (CEO Shina Chung) announced that the company will participate in the 2025 Culture, Sports and Tourism AI and Digital Innovation Forum hosted by the Korea Culture Information Service Agency.

 

The forum is organized by the Ministry of Culture, Sports and Tourism and hosted by the Korea Culture Information Service Agency, with the support of the Korea Sports Promotion Foundation, the Korea Tourism Organization, the Arts Council Korea, the Korea Copyright Commission, and the Korea Creative Content Agency. Under the theme “Digital Transformation in Everyday Culture, Innovation Led by AI,” the event serves as a platform for sharing AI and digital transformation cases across a wide range of fields, including culture, arts, content, sports, tourism, and copyright.

 

Kakao will take part in a session on open-source AI technologies hosted by the Korea Copyright Commission, where it will present its in-house AI model development strategy for Kanana and its broader open-source initiatives.

 

The presentation will be delivered by Kanana Performance Lead Byung-hak Kim. He will discuss the importance of AI models that demonstrate a deep understanding of Korean sentiment and culture, and introduce the lineup of Kakao’s proprietary AI model, Kanana. Designed to deliver high-performance models tailored for AI services, Kanana has undergone multiple rounds of refinement and is currently preparing to develop Kanana-2, which further strengthens its efficient inference architecture. The presentation will highlight the key characteristics of Kanana-2, including the application of Multi-Head Latent Attention (MLA) and a Mixture of Experts (MoE) architecture, as well as plans to advance lightweight models and develop multimodal language models capable of processing diverse forms of information.

 

Kakao has been steadily releasing Kanana models as open source to lower barriers to AI technology and enable broader participation in innovation. Since February, the company has open-sourced language models, multimodal language models, and guardrail models, achieving more than 400,000 cumulative downloads. Among the released models, Kanana-1.5-8b ranked first on the Ko-agent Bench leaderboard, which evaluates agent capabilities optimized for real-world use cases, earning recognition for performance that rivals or exceeds that of leading global models. Kakao has also applied commercial-use licenses to its models, strengthening Korea’s AI competitiveness and providing a foundation for freely experimenting with and deploying domestic AI models through open technology sharing.

 

Performance Lead Byung-hak Kim commented, “Beyond applying AI research outcomes to services, Kakao is committed to developing and expanding models that uphold social value and responsibility, with a strong focus on transparency and accessibility. By sharing the benefits of AI technology with society as a whole rather than confining them to a single company, we aim to continuously contribute to the revitalization of the domestic AI ecosystem through inclusive innovation.”

 

Meanwhile, Kakao is jointly running an AI Agent Competition with the Korean Institute of Information Scientists and Engineers. Undergraduate and graduate students are eligible to apply in teams of up to three members. Teams will receive bonus points if they utilize Kakao’s open MCP platform, PlayMCP, or incorporate Kakao’s Kanana AI models released on Hugging Face as core logic. Mentoring by Kakao employees for finalist teams will begin on November 24, with an awards ceremony scheduled for December 17. A total prize pool of KRW 20 million will be awarded to 10 teams, with the grand prize winner receiving KRW 6 million. (E.O.D.)