AI BriefingKrafton Releases Open-Source Preview of Its In-House 30B-Class LLM 'Raon-1 mini'
·2026.10.21 14:28
Key point
The Raon-1 mini preview, a 30B-class MoE model, has been released under the Apache 2.0 license.
Details
- Krafton has open-sourced the preview of its in-house developed 30B-class LLM, 'Raon-1 mini', on Hugging Face under the Apache 2.0 license
- Combines an MoE architecture that activates only 4B of the 30B total parameters with its proprietary linear attention technology to achieve fast response times and low inference costs
- Demonstrated superior performance compared to overseas models of similar scale, recording levels approaching those of major domestic models on some metrics
- Features strengths in natural Korean conversation, instruction-following capabilities such as persona settings, and agent execution performance; the official version and technical report are scheduled for future release
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Janas-LLM Streams MoE Experts from NVMe SSD to Run Models Larger Than RAM