KRAFTON Open-Sources Raon-1 nano: 1.5B Effective Params, 262K Context, On-Device Ready
pmttyji · reddit · 2026-11-12
KRAFTON AI released Raon-1 nano under Apache 2.0: a dense model with 1.5B effective parameters (3.1B total including a 1.5B per-layer embedding table), strong Korean/English conversation and tool calling, and 262,144-token context. It uses a hybrid layout of 18 Raon Delta Attention linear-attention layers plus 6 Gated MLA layers (7 KiB/token KV cache), Raon-mHC, NoPE, and a 55M MTP module. Derived from a 30B-A4B MoE via MoE-to-Dense pruning and distillation, with 3T tokens of continued pre-training plus SFT/OPD/RL. Available in GGUF, MLX, and LiteRT-LM for on-device inference including Apple silicon.
Related event: Krafton Open-Sources Raon-1 nano, an On-Device 1.5B Model Beating Peers(2 posts)→
More from Models
- Google engineer spotlights AI Studio redesign, name-drops unreleased Gemini 4 Carbon — ammaar · 2026-11-12
- All frontier models cheat on HoneyBench v0.2, with cheat rates from 12% to 86% — ChowdhuryNeil · 2026-11-12
- Google DeepMind Code Summit talk urges devs to prepare for massive inference speed gains — altryne · 2026-11-12
- Never letting the model write directly: custom tools make Fable 5.2 prose feel human — Flomerboy · 2026-11-12
- Vals AI analyzed 200K benchmark runs: 14 models swear in reasoning, Xiaomi MiMo V2.6 Pro is foulest — ziqiao_ma · 2026-11-12
- Decido-1: open 26B MoE decision model beats Jev 1.13, runs CPU-only on laptops — gauravapiscean · 2026-11-12