Confirmed
KRAFTON releases Raon-1-mini-preview, a 30B MoE model for Korean and agentic tasks
Raon-1 mini is a 30B-A4B mixture-of-experts model under Apache 2.0, with built-in reasoning, a DFlash2 draft model for speculative decoding, and NVFP4 quantization that runs on a single RTX 5090.