Kakao has developed 'Kanana-O', an on-device multimodal AI model that can operate directly on devices such as smartphones. It is expected to be applied to core AI services like 'Kanana Space' within KakaoTalk. Following the vision encoder, Kakao plans to release an audio encoder within the year. Attention is growing as Kakao accelerates the era of multimodal AI that processes voice input and image understanding beyond text.
According to industry sources on the 25th, Kakao is continuing to develop and refine the on-device Kanana-O model. Kanana-O is an omni-model that processes text, images, and voice, and Kakao has already secured a version that can run on smartphones.
The technology behind Kanana-O is expected to be applied to Kakao's AI services, such as Kanana Space, in the long term. Kanana Space is an AI service where the character 'Ryan' appears when swiping down with two fingers on the KakaoTalk screen, receiving user requests via voice and text. Kakao plans to start a closed beta test (CBT) in December.
Initially, Kakao will operate a server-based model for Kanana Space. However, in the long term, it plans to apply the on-device model, processing simple tasks on the smartphone and complex tasks via the Frontier model to enhance efficiency.
On-device AI technology, which operates AI models directly on devices like smartphones, reduces server dependency and improves response speed. It also minimizes the transmission of user data to external servers, enhancing privacy and security.
At 'if(kakao)26' on the 13th, Noh Byung-seok explained, “Requests that can be processed on the device are handled first on the user's smartphone before being sent to the server. This allows for faster responses and reduces the scope of sensitive information being transmitted externally.”
Kakao plans to strengthen its multimodal competitiveness by continuously developing AI models and supporting encoders. On the 21st, Kakao open-sourced its self-developed vision encoder. It also plans to release a speech encoder as open source within the year. The goal is to support researchers and companies in utilizing AI core technologies like vision encoders and developing services based on various self-developed AI models.
A Kakao representative stated, “Going forward, we will enhance both the performance and efficiency of our models based on an understanding of the Korean language and user context. We will also advance the foundational technology for AI agent services while continuously sharing research outcomes.”
