The neural-network architecture, built around attention, that underlies virtually all modern AI models.
Broad technical consensus on what this term means. Disagreements that invoke it are about the world, not the word.
The transformer (introduced in 2017's "Attention Is All You Need") arranges attention and standard network layers so that a whole sequence can be processed in parallel rather than word by word. That parallelism is what let training scale onto huge GPU clusters, and that scalability, more than any single insight, opened the current era.
Nearly every prominent model, GPT ("generative pre-trained transformer"), Claude, Gemini, Llama, is a transformer. When people debate whether progress requires "a new architecture," the transformer is the incumbent they mean.