LGAI-EXAONE's large mixture-of-experts model, activating 23 billion parameters out of 236 billion total — giving it the computational efficiency of a smaller model while drawing on a much larger knowledge base. It handles long-context tasks well, with a 262K token window that lets it work through lengthy documents or codebases without losing track. As an open-weight release, it can be run locally, though the full model's size demands serious hardware.