Raon — local AI model by KRAFTON AI
Written by Jakub Rusinowski · Last updated
KRAFTON's from-scratch model family. Raon-1 mini preview is a 28.8B mixture-of-experts with about 4.3B active parameters. 18 of its 25 layers use Raon Delta Attention with a fixed-size state, and only the 7 MLA layers keep a KV cache. It has a 262K context, thinking mode and tool calling, and targets Korean conversation and agent tasks. Apache 2.0, labelled a preview whose specs may change before the final release.
Variants
The smallest Raon variant needs about 20 GB of VRAM at Q4_K_M — quantized weights plus framework overhead, before any KV cache.
| Model | VRAM at Q4 | VRAM | Context | Run it |
|---|---|---|---|---|
| Raon-1 mini preview 30B-A4B → 28.8B (4.3B active) | ~20.5 GB | 262,144 | cloud-hosted tag |
Memory is quantized weights plus overhead at Q4_K_M, from the same engine as the GPU & VRAM checker.
How to run Raon locally
Install Ollama, then pull the tag.
No Ollama tag. KRAFTON documents Transformers, vLLM, MLX, and llama.cpp through its own fork (github.com/krafton-ai/llama.cpp) for the official GGUF files.
Pick a size above for its own VRAM figure, speed estimate and install command.
Licence
Commercial use permitted. No usage restrictions beyond attribution.
Applies to: Raon-1 mini preview 30B-A4BRecommended GPU
The cheapest catalogued GPU that runs Raon locally (min 20 GB VRAM) is the AMD Radeon RX 7900 XTX (24 GB).
Raon — frequently asked questions
How much VRAM does Raon need?
Raon needs about 20 GB VRAM at Q4_K_M quantization for its smallest variant. Variants: Raon-1 mini preview 30B-A4B (20 GB, Q4_K_M). On Apple Silicon, unified memory counts toward this requirement.
Can I run Raon on an RTX 4090 (24 GB)?
Yes — Raon runs on an RTX 4090 (24 GB) and other 24 GB cards such as the RTX 3090. Smaller variants also fit comfortably on 8–16 GB GPUs at Q4_K_M.
What quantization should I use for Raon?
Q4_K_M is the best balance of quality and VRAM for Raon in most cases. Choose Q8_0 for near-lossless quality if you have spare VRAM, or smaller quants (Q3/Q2) only when memory is tight.
How do I run Raon with Ollama?
Raon has no Ollama tag yet. No Ollama tag. KRAFTON documents Transformers, vLLM, MLX, and llama.cpp through its own fork (github.com/krafton-ai/llama.cpp) for the official GGUF files.