Krafton Launches In-House LLM 'Raon-1 Mini' Targeting Korean Language
Krafton has introduced its own large language model, Raon-1 mini, designed to enhance Korean language processing and agent performance.
Krafton has introduced its own large language model, Raon-1 mini, designed to enhance Korean language processing and agent performance.
Krafton has unveiled its self-developed large language model (LLM), 'Raon-1 mini', aimed at improving Korean language processing and agent performance while ensuring cost efficiency.
The Raon-1 mini model employs a unique architecture that activates only 4 billion out of a total of 30 billion parameters, optimizing both performance and cost. Krafton claims that this model surpasses comparable global models in reasoning and Korean conversation tasks.
In performance evaluations, Raon-1 mini achieved an impressive score of 84.3 on the GPQA Diamond benchmark, outperforming other models like Qwen3.6-35B-A3B and Gemma 4 26B-A4B. It also excelled in mathematical reasoning and user instruction adherence, scoring 93.3 and 84.2 respectively.
Krafton plans to use Raon-1 mini as a foundational model for future AI developments, emphasizing its potential in various industries. Chief AI Officer Kang Woo-kyung highlighted the model's competitive edge, stating, "This is the result of independently developing a competitive language model through our AI research and development capabilities accumulated over more than five years."
