Transformer‑based language model trained to predict next tokens; adaptable via prompting or fine‑tuning.