A large language model (LLM) is an AI system trained on huge amounts of text to predict the next piece of text in a sequence. By doing that prediction very well, it learns grammar, facts, styles and patterns of reasoning, which lets it answer questions, write, summarize, translate and code. GPT (behind ChatGPT), Claude, Gemini, Llama and DeepSeek are all LLMs.
How a large language model works
An LLM reads text as tokens, small chunks of words, and turns each one into a list of numbers. Layers of a neural network called a transformer compare every token with every other token to work out context, a mechanism known as attention. At the end, the model outputs a probability for every possible next token, picks one, adds it to the text and repeats. A full answer is that loop running hundreds of times.
How LLMs are trained
Training happens in stages:
- Pretraining: the model reads trillions of tokens from books, websites, code and other text and learns to predict the next token. This is where most knowledge comes from and where most of the cost is.
- Fine-tuning: the model is trained on examples of good instructions and answers, so it behaves like an assistant instead of just continuing text.
- Feedback training: people or other models rate answers, and the model is adjusted to prefer the better ones. This is often called reinforcement learning from human feedback (RLHF).
- Reasoning training: newer models are also trained to work through problems step by step before answering, which improves math and coding.
What “large” means
“Large” refers to the number of parameters, the internal values set during training, and the amount of data. Modern models have billions to over a trillion parameters. More parameters usually means more capability, but also more computing power and cost to run, which is why providers offer small, fast models alongside flagship ones.
What LLMs can do well
LLMs are good at drafting and editing text, summarizing long documents, translating, explaining concepts at different levels, writing and fixing code, extracting data from messy text, and brainstorming. With tools added, such as web search or a code interpreter, they can also look up current information and run calculations.
What LLMs get wrong
An LLM generates plausible text, not verified truth, so it can state false things confidently. These errors are called hallucinations. Models also have a knowledge cutoff, can be inconsistent between runs, and struggle with exact counting or long arithmetic unless they use a tool. They are only as good as the context you give them, which is why good prompts matter.
LLMs vs chatbots
The LLM is the engine; the chatbot is the car. ChatGPT, Claude.ai and the Gemini app add memory, file upload, voice, search and safety filters on top of the model. That is why the same model can feel different in different apps. To see how this works in one product, read how ChatGPT works.
Major LLMs in 2026
The best-known families are GPT from OpenAI, Claude from Anthropic, Gemini from Google, Grok from xAI, DeepSeek, and open-weight models such as Meta’s Llama, Alibaba’s Qwen and Mistral. Our LLM comparison shows how they differ in practice.
Related questions
If you want to go deeper, see what AI tokens are, why AI hallucinates, and whether AI detectors can tell LLM text from human writing.
Frequently asked questions
What does LLM stand for?
LLM stands for large language model: an AI model trained on very large amounts of text to understand and generate language.
Is ChatGPT an LLM?
ChatGPT is an app built on OpenAI's GPT large language models. The LLM generates the text; the app adds features like chat history, web search, file upload and image generation.
What makes a language model large?
The number of parameters, the adjustable values learned in training, usually billions, and the amount of training data, usually trillions of tokens.