Read original
HuggingFace Daily PapersEunbi ChoiModels91

K-EXAONE 2.0 Technical Report: A 750B-Parameter Open-Weight Multilingual MoE Model

Original title:K-EXAONE 2.0 Technical Report

LG AI Research presents K-EXAONE 2.0, an open-weight multilingual foundation model built by upcycling K-EXAONE rather than training from scratch. The model uses a Mixture-of-Experts architecture with 750B total parameters and approximately 37B activated per token, supports contexts up to 256K tokens, and expands language coverage from six to ten languages. Its training pipeline combines continual pre-training, difficulty-focused mid-training, and post-training for reasoning, agentic coding, multilingual performance, and safety in Korean sociocultural contexts. It is released under Apache 2.0.

Why it's worth reading

The combination of a 750B MoE, 256K context, and Apache 2.0 release gives developers an immediately relevant reference for long-context, coding-agent, and multilingual deployment decisions.

Tags

K-EXAONELG AI ResearchMoE开放权重256K上下文多语模型代码代理模型安全