KRAFTON AI

The AI organization at KRAFTON. We research and build deep learning systems for game creation, game AI, and human–AI interaction — from foundational research to real deployment in products and internal tools.

Latest release

Raon-1 nano — an on-device open-weight LLM (Apache 2.0) with 1.5B effective and 3.1B total parameters, derived from Raon-1 mini preview through MoE-to-Dense pruning and distillation. Delivers natural Korean and English conversations, strong contextual understanding, reliable instruction following, and effective tool use.

Raon-1 mini preview — a 30B-A4B open-weight LLM (Apache 2.0) with an MoE + linear-attention architecture. Deep Korean, fluent English, strong agentic capabilities. Leads similar-size open MoE models on science reasoning, instruction following, agentic coding, search, and Korean conversation.

Playground · Model site & free API · Docs

Raon model family

• Raon-1 nano — E1.5B open-weight LLM · Model card
• Raon-1 mini — 30B-A4B open-weight LLM · Model card · Playground
• A.X-K2 Raon Speech — 21B-A3B speech model · Model card · Playground