14B (dense)
Apache 2.0
First open-source HyperCLOVA X reasoning model with advanced AI agent capabilities; trained at 1% cost of comparable global models (52.6× lower than Qwen2.5-14B, 91.38× lower than Qwen3-14B) through parameter pruning and knowledge distillation; multi-stage RL pipeline: SFT → RLVR (Reinforcement Learning with Verifiable Rewards) → Length Controllability → RLHF; solves complex problems step-by-step, selects appropriate tools/functions, and self-corrects errors; superior Korean language and cultural understanding vs Qwen3-14B while competitive in math/coding; outperforms some 32B models in evaluations; custom LLM architecture based on LLaMA with μP and Peri-LN techniques; optional thinking mode for controllable reasoning; demonstrates that efficiency innovations (pruning, distillation, targeted RL) can rival brute-force scaling; commercial open-source enabling Korean AI ecosystem development; >500k downloads.