Kanana-o is presented as South Korea's first integrated multimodal language model, developed in May 2025, capable of perceiving, listening, and comprehending like a human, and expressing rich emotions. It is positioned as a pioneering, general-purpose AI designed to overcome task-specific limitations.

The core methodology of Kanana-o revolves around its sophisticated integrated multimodal architecture. It simultaneously processes and understands information from diverse modalities, including text, images, and speech. This capability allows the model to deeply interpret complex intentions and situations, particularly within the nuances of the Korean language and its cultural context. Technologically, this implies a unified deep learning framework, likely leveraging advanced transformer-based architectures, that can learn cross-modal representations. The model is engineered for advanced instruction following and aims to provide practical utility beyond mere benchmark performance.

Key features of Kanana-o include:

1. Deep Understanding: It can simultaneously understand and process text, images, and voice, uniquely capable of interpreting intricate intentions and contextual nuances inherent in the Korean language.
2. Natural Speech Generation: The model generates human-like speech with rich emotional expression, considering elements such as intonation, speaking speed, emotion, and speaker characteristics. This includes precise pronunciation, clear audio quality, and natural Korean 발화 (utterance). Its speech generation capabilities extend to supporting podcast-style speech, multi-turn conversational scenarios, and multi-speaker dialogue synthesis (TTS), as well as video sound production.
3. Versatility (General-Purpose Multimodal Model): Kanana-o is not confined to specific tasks but offers broad support for various real-world use cases, indicating a robust and adaptable underlying architecture.

Kanana-o is being introduced through a Closed Beta Test via API access, with applications open until May 27, 2026. The purpose of this beta program is to gather user feedback and assess the stability and performance of the new model in real-world scenarios. While free to experience, the beta service has usage limits and potential variability in service stability. Selection for the beta prioritizes users who possess concrete utilization scenarios, technical implementation capabilities, and a commitment to providing active feedback, aiming to cultivate an early community of developers who can create substantial value with the model.