PulseAugur
EN
LIVE 00:33:56

LG AI Research releases K-EXAONE 2.0 with 750B parameters, 256K context

By PulseAugur Editorial · [2 sources] ·

LG AI Research has released K-EXAONE 2.0, an open-weight multilingual foundation model. This updated Mixture-of-Experts model boasts 750 billion total parameters, with approximately 37 billion activated per token, significantly increasing its capacity over its predecessor. K-EXAONE 2.0 supports a 256K token context length, expands multilingual capabilities to ten languages, and shows notable improvements in agentic coding and long-context understanding. The model is released under the Apache 2.0 license, encouraging broader ecosystem development. AI

IMPACT This release provides a powerful open-weight foundation model for researchers and developers, potentially accelerating advancements in multilingual AI and long-context understanding.

RANK_REASON Technical report detailing a new open-weight model release.

Read on Hugging Face Daily Papers →

AI-generated summary · Google Gemini · from 2 sources. How we write summaries →

LG AI Research releases K-EXAONE 2.0 with 750B parameters, 256K context

COVERAGE [2]

  1. arXiv cs.CL TIER_1 English(EN) · Eunbi Choi, Kibong Choi, Sehyun Chun, Seokhee Hong, Junwon Hwang, Hyojin Jeon, Ahra Jo, Hyunjik Jo, Yeonsik Jo, Minhyeok Jung, Doyoung Kim, Heegyu Kim, Joonkee Kim, Seonghwan Kim, Soyeon Kim, Sunkyoung Kim, Yireun Kim, Yongil Kim, Byungoh Ko, Changhun Le… ·

    K-EXAONE 2.0 Technical Report

    arXiv:2608.04505v1 Announce Type: new Abstract: This technical report presents K-EXAONE 2.0, an open-weight multilingual foundation model developed by LG AI Research as a step in our effort toward global frontier-scale foundation models. Rather than training from scratch, we upcy…

  2. Hugging Face Daily Papers TIER_1 English(EN) ·

    K-EXAONE 2.0 Technical Report

    This technical report presents K-EXAONE 2.0, an open-weight multilingual foundation model developed by LG AI Research as a step in our effort toward global frontier-scale foundation models. Rather than training from scratch, we upcycle K-EXAONE and expand its architecture, yieldi…

RELATED ENTITIES

RELATED TOPICS