Decoder
A model, or part of one, that writes text one token at a time, each token seeing only those before it.
Also: decoder-only model
A decoder uses masked self-attention so that every position can see only earlier positions, which lets it be trained to predict the next token and then used to generate text. GPT-1 in 2018 showed that a decoder-only transformer trained on plenty of text could be adapted to many tasks, and since GPT-3 almost every large general model, including Llama, Claude, Gemini, Qwen and DeepSeek, is decoder-only. The design is simple and scales well, and the same stack handles both reading the prompt and writing the reply.