K EXAONE

By LG AI

New Text Gen 7

Released: December 31, 2025

Overview

K-EXAONE is LG AI Research’s 236B-parameter MoE language model with 23B active experts, 256K context and strong multilingual reasoning, built as a national Korean foundation model with efficient hybrid attention.

Description

K-EXAONE extends the EXAONE 4.0 line into a sovereign foundation model, using a mixture-of-experts design plus hybrid global-local attention to cut memory and computation for very long inputs. It supports Korean, English, Spanish, German, Japanese, Vietnamese and more, and targets reasoning, agentic workflows and long-context applications like document copilots, with open weights and temporary free API access.