What makes a language model large?
Mostly the number of parameters, the internal numbers learned in training, which runs from a few billion to well over a trillion, plus the huge training data. There is no exact size cutoff for the label.
An AI model trained on vast amounts of text to predict the next token, which lets it write, answer questions and code.
A large language model, or LLM, is an AI system trained on enormous amounts of text, and often code and images, to predict what comes next in a sequence. That simple training goal, repeated across trillions of tokens, produces a model that can write, summarize, translate, answer questions and write code. It is large because it has billions of internal numbers, called parameters, adjusted during training. After the base training, most chat models get further training to follow instructions and to be helpful and safe. Claude, GPT and Gemini are LLMs, as are open-weight families such as Llama, Qwen, DeepSeek and Mistral.
Knowing that an LLM predicts likely text explains a lot of its behavior: why it is fluent, why it can be confidently wrong, why the information you give it in the prompt matters so much, and why the same question can get different answers. It also explains why tools, tests and real documents make LLMs far more reliable at practical work.
Given the start "def add(a, b):", an LLM predicts that the next tokens are very likely "return a + b". The same mechanism, scaled up and guided by instructions, writes a whole function, explains a stack trace or drafts an AGENTS.md file from a description of your stack. Nothing in that process checks the result against reality, which is why tests and real sources matter.
An LLM is not a database or a search engine. It does not look up stored facts; it generates text that fits patterns it learned. Connect it to real sources when facts matter.
All 28 terms are in the AI glossary. For models side by side, see frontier vs open-weight.
Mostly the number of parameters, the internal numbers learned in training, which runs from a few billion to well over a trillion, plus the huge training data. There is no exact size cutoff for the label.
That is debated. They clearly capture a lot of structure about language and the world, but they work by predicting text, without direct experience, and can fail in ways a person who understood would not.
ChatGPT is an app built on OpenAI GPT models, which are LLMs. Similarly, the Claude app uses Anthropic Claude models and the Gemini app uses Google Gemini models.