BERT is a word-level model, while Sentence-BERT (SBERT) is a sentence-level model, designed to generate sentence 1embeddings. 

BERTSBERT
How it processes textInterprets words one at a timeEmbeds entire sentences into vectors
ArchitectureUses a cross-encoderUses a bi-encoder
How it’s usedGood for understanding word nuancesGood for semantic similarity tasks