KRAFTON's Raon-1 Mini Language Model Matches Larger Competitors on Reasoning, Sets New Benchmark for Agentic Tasks

KRAFTON has released a preview of Raon-1 Mini, a 30-billion-parameter mixture-of-experts language model trained entirely in-house that demonstrates competitive performance with larger models while requiring significantly fewer active parameters during inference. The Korean game developer's new model activates only 4 billion of its 30 billion total parameters per token, substantially reducing computational requirements compared to similarly-sized competitors. The architecture combines 18 linear attention layers with 7 full attention layers—a hybrid design incorporating KRAFTON's proprietary Raon Delta Attention technology that enables context windows up to 262,144 tokens with minimal memory overhead.