Which of the following statements best categorizes the advantages of the Transformer architecture compared to traditional RNNs in natural language processing tasks?

Transformers use self‑attention, letting the model look at every word at once, so they can be trained in parallel and quickly handle sentences of any length. This ability lets them capture long‑range relationships that RNNs, which read one token at a time, often miss or blur. Because all tokens are processed together, Transformers train faster and scale better to large datasets. For example, a transformer can instantly compare the first and last words of a long paragraph, while an RNN would have to walk through every intermediate word to make that connection. As a result, transformers usually give higher accuracy and faster inference on NLP tasks.