LG AI Research has introduced K-EXAONE 2.0, an open-weight multilingual foundation model designed to push the boundaries of global frontier-scale AI. Rather than training from scratch, the team upcycled the previous K-EXAONE model and expanded its architecture, resulting in a Mixture-of-Experts (MoE) model with 750 billion total parameters and approximately 37 billion activated per token—more than three times the capacity of its predecessor.
The model supports context lengths of up to 256K tokens, a significant upgrade for long-context tasks, and expands multilingual coverage from six to ten languages. Its training pipeline combines continual pre-training, difficulty-focused mid-training, and post-training to strengthen reasoning, agentic coding, multilingual capability, and safety grounded in Korean sociocultural contexts.
Across nine evaluation categories reflecting practical use, K-EXAONE 2.0 shows improvements over its predecessor and remains competitive with other open-weight models. The largest gains are seen in agentic coding and long-context understanding, with particular strengths in long-context retrieval and safety.
Released under the Apache 2.0 license, K-EXAONE 2.0 is available for the wider AI ecosystem to evaluate, deploy, adapt, and build upon. The team describes this release as the beginning of their challenge toward the global frontier, not the endpoint.