A language model is an AI model that has learned statistical patterns of language from large amounts of text and uses them to generate, summarize, translate or classify text. Large language models (LLMs) with billions of parameters underpin chat assistants, coding aids and AI agents. Technically, they predict the most probable next fragment of a word, one step at a time.
Under the AI Act, large language models usually qualify as general-purpose AI models. Their providers carry their own obligations under Chapter V, such as technical documentation and a summary of the content used for training. Organizations that build a language model into their own applications through an interface review the contract, data use and storage location. Because the model does not check facts, expert review of the results remains necessary, see T13 Hallucination / faulty output.