A focused specialist that excels at capturing semantic similarity between sentences, even when they're phrased differently. Built on MPNet's bidirectional attention, it produces dense 768-dimensional embeddings that cluster paraphrases tightly together. It handles everyday language well but wasn't designed for highly technical or domain-specific text.