Definition
A large language model (LLM) is an artificial intelligence system trained on enormous amounts of text to predict the next word in a sequence. That simple objective, applied at massive scale, produces a model that can write, summarise, translate, answer questions, follow instructions and write code. A large language model does not look things up in a database. It generates text based on patterns learned during training, plus whatever you put in front of it in the prompt.
In a company, a large language model is the engine behind AI assistants, customer support bots, document analysis, contract review, sales email drafting and internal knowledge search. You access it through a chat product, through an API your developers call, or embedded in tools you already use. The model is generic; the value comes from connecting it to your data, your rules and your workflows, usually through retrieval and tool use.
In 2026 the frontier is held by a small number of labs, with the Claude, GPT and Gemini families at the top and strong open-weight alternatives such as Qwen, DeepSeek, Llama and Mistral close behind. Prices per token have fallen sharply year after year. The main misconception is that a large language model is a knowledge base. It is a reasoning and writing engine that needs grounding in your facts to be reliable.
In practice
A B2B software company connects a large language model to its help centre and past tickets. It now drafts the first answer to every support request, and agents approve or edit rather than write from scratch.
Why it matters
LLMs make language, the raw material of most office work, programmable. In my companies, they changed the cost structure of support, sales content and internal documentation within months.
Frequently asked questions
- What is the difference between an LLM and ChatGPT?
- ChatGPT is a product; the large language model is the engine inside it. The same distinction applies to Claude and Gemini. Companies can use the underlying models directly through an API to build their own tools, without the consumer chat interface.
- Can a large language model use my company data?
- Yes, in two main ways. You can include relevant documents in the prompt at query time, usually through retrieval-augmented generation, or you can fine-tune a model on your data. For most businesses, retrieval is cheaper, safer and easier to keep up to date.