A quiet, focused specialist that converts text into dense numerical representations. It excels at capturing semantic similarity — sentences with related meanings end up close together in its embedding space. The 514-token context ceiling means it works best with short passages rather than long documents, and it produces no generated text, only vectors.