Open-Source Release

Kakao disclosed the two Kanana 2 Embedding models through the developer platform Hugging Face, where the models, their documentation and related assets including the Kanana logo are hosted in public repositories. The release means developers and researchers can freely access the embedding models rather than relying solely on Kakao's internal services.

Embedding models are a core component of AI search and retrieval systems. They convert text into numerical vectors so that semantically similar documents can be found efficiently, a function widely used in retrieval-augmented generation (RAG) and search applications. By open-sourcing these models, Kakao positions Kanana 2 Embedding as a resource for the broader developer community building Korean-language search and AI retrieval solutions.

Kakao's AI Model Strategy

The Kanana series represents Kakao's push to develop its own AI models for Korean-language use cases. The embedding models complement larger language model efforts by handling the retrieval side of search-driven AI services, where finding relevant source material quickly and accurately is critical for answer quality.

The release follows a broader industry trend in which Korean tech companies publish smaller, specialized models openly while continuing to develop proprietary systems for their consumer services. Open embedding models in particular allow startups and independent developers to build retrieval pipelines without licensing costs, which can accelerate adoption of domestic AI tooling.

The models are now downloadable via Kakao's official Hugging Face repositories, and the company has not announced restrictions beyond the terms attached to the published repositories. The claim that Kakao has released two AI search models, Kanana 2 Embedding, as open source is True.