{"version":1,"type":"story","url":"https://digestai.news/story/krafton-releases-raon-1-mini-preview-a-30b-a4b-moe-model-with-262-144","json":"https://digestai.news/story/krafton-releases-raon-1-mini-preview-a-30b-a4b-moe-model-with-262-144.json","markdown":"https://digestai.news/story/krafton-releases-raon-1-mini-preview-a-30b-a4b-moe-model-with-262-144.md","slug":"krafton-releases-raon-1-mini-preview-a-30b-a4b-moe-model-with-262-144","headline":"Krafton releases Raon-1 mini preview, a 30B-A4B MoE model with 262,144 token context","summary":"KRAFTON AI announced a preview of its Raon-1 mini language model, a 30B-A4B Mixture‑of‑Experts system built from scratch. The model uses 18 linear attention layers and 7 full attention layers, and it supports up to 262,144 tokens of context. It runs on a single NVIDIA RTX 5090 GPU (32 GB VRAM) and offers quantization in MXFP8 and NVFP4 formats.\n\nRaon-1 mini’s sparse MoE activates six of 128 routed experts plus one shared expert per token, using 4.3B of the 28.8B total parameters. The preview includes a DFlash2 draft model for speculative decoding, delivering up to 2.18× higher output throughput on a single NVIDIA B300 compared with autoregressive decoding. The model is released under the Apache 2.0 license, with open‑weight weights, and can be accessed via a free limited‑time playground or through hosted API access after agreeing to separate Hosted Access Terms. The release is labeled a preview, and the article notes that specs and results may change before the final version.","keyPoints":["Raon-1 mini is a 30B-A4B MoE model with 18 linear and 7 full attention layers","Supports up to 262,144 tokens of context and runs on a single NVIDIA RTX 5090 (32 GB VRAM)","Sparse MoE activates 6 of 128 experts plus 1 shared, using 4.3B of the 28.8B parameters per token"],"whyItMatters":"A long‑context, multilingual open‑weight model gives developers a free alternative for Korean‑centric AI applications and large‑scale reasoning.","category":{"slug":"models","name":"Generative AI & Models","url":"https://digestai.news/category/models"},"entities":{"companies":["KRAFTON AI"],"models":["Raon-1 mini"],"people":[]},"firstPublishedAt":"2026-10-21T07:54:08Z","updatedAt":"2026-10-21T07:54:08Z","sourceCount":1,"hasPrimarySource":true,"sources":[{"outlet":"huggingface.co","title":"KRAFTON/Raon-1-mini-preview · Hugging Face","url":"https://huggingface.co/KRAFTON/Raon-1-mini-preview","publishedAt":"2026-10-21T07:54:08Z","type":"primary","primary":true,"lead":true}],"sourceNotes":null,"discussions":[{"site":"Reddit","url":"https://www.reddit.com/r/LocalLLaMA/comments/1xbh0ev/kraftonraon1minipreview_hugging_face/","points":null}],"thread":null,"cite":{"text":"Digest AI, \"Krafton releases Raon-1 mini preview, a 30B-A4B MoE model with 262,144 token context\", 21 October 2026, https://digestai.news/story/krafton-releases-raon-1-mini-preview-a-30b-a4b-moe-model-with-262-144","publisher":"Digest AI","title":"Krafton releases Raon-1 mini preview, a 30B-A4B MoE model with 262,144 token context","datePublished":"2026-10-21T07:54:08Z","url":"https://digestai.news/story/krafton-releases-raon-1-mini-preview-a-30b-a4b-moe-model-with-262-144"},"generatedBy":"Written by Digest AI's editorial model from the linked sources; the sources are the record.","license":"Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse"}