PUBG Maker KRAFTON Releases Open Raon-1 Nano: 30B MoE Distilled into a 3.1B Dense Model
Kangwook_Lee · x · 2026-11-12
KRAFTON, the company behind PUBG, released Raon-1 nano with downloadable weights under Apache 2.0, supporting English and Korean. They pruned and distilled the Raon-1 mini preview into a dense model, continued training on 3 trillion tokens, then post-trained. Total 3.1B parameters, described as an effective 1.5B dense model.
Reported benchmarks vs. Qwen3.5 2B, Gemma 4 E2B, MiniCPM5 2B, LFM3 1.7B:
- IFBench: 75.9 (leading)
- PersonaMem v2 32K: 38.8 (leading)
- AndroidControl: 36.0 (leading)
- BrowseComp Plus: 34.5 (behind MiniCPM5's 37.2)
Positioned as a small assistant for conversation, instruction following and tool use without a giant model behind every interaction.
Related event: KRAFTON open-sources 1.5B on-device model Raon-1 nano(4 posts)→
More from Models
- Qwen 3.8 27B on a laptop can read model weights, replicating the Weight-Space Programming paper — dl_weekly · 2026-11-12
- KOL predicts Gemini will be the top frontier LLM by year-end as rivals stumble — iruletheworldmo · 2026-11-12
- Dense rewards lose to standard GRPO at matched compute, arXiv paper finds — burny_tech · 2026-11-12
- Dev builds npx driftcost CLI to check if Claude Code token usage silently shifted between Opus versions — Tough_Stay8909 · 2026-11-12
- Reflection's Beam free until Nov 12; fx is a 5.97MiB Zig-native open coding agent — evilrabbit_ · 2026-11-12
- Community speculates Google may announce Gemini 4 (Krypton) instead of Gemini 4 Flash — algo_diver · 2026-11-12