Large Language Model (LLM)

A large language model (LLM) is a type of artificial-intelligence model trained on very large amounts of text to predict and generate human language. Built on the transformer neural-network architecture, an LLM learns statistical patterns of language and can produce coherent text, answer questions, summarize documents, translate, write code, and follow instructions. Well-known families include Anthropic's Claude, OpenAI's GPT models, and models served through providers such as AWS Bedrock.

LLMs work by predicting the next token (a word or word-piece) given the preceding context, repeated to build a full response. Their behavior is shaped by training data, a knowledge cutoff date, and a context window that limits how much text they can consider at once. Practical limitations include the potential to generate confident but incorrect statements, sensitivity to how prompts are phrased, and per-token usage costs charged by model providers.

LLMs are the reasoning engine behind most current AI agents and assistants; techniques like retrieval-augmented generation and tool use extend them with current and private data and the ability to take actions. In products that let customers bring their own provider key (BYOK), the LLM is supplied by the customer's chosen vendor, so model choice, cost, and data handling stay under the customer's control.