A large language model (LLM) is a neural network trained on vast amounts of text to do one deceptively simple thing: predict the next token in a sequence. Everything an LLM appears to do — answer questions, write code, summarize documents — is that prediction loop running one token at a time, each choice informed by patterns learned from trillions of words. ChatGPT, Claude and Gemini are all LLMs with product layers on top.
How it works, in one paragraph
Nearly every modern LLM is a transformer, an architecture whose attention mechanism lets the model weigh every part of the input against every other part — so “bank” near “river” reads differently than “bank” near “loan”. Training happens in two phases: pretraining, where the model learns language, facts and reasoning patterns by predicting held-out text at enormous scale, and post-training (instruction tuning, RLHF), where the raw text predictor is shaped into a helpful assistant that follows instructions and declines harmful requests.
Why “large” matters
Scale is the ingredient that made LLMs interesting. As parameter counts and training data grew, capabilities appeared that smaller models simply didn’t have — multi-step reasoning, working code, tool use. By 2026 the frontier spans closed models like Claude Fable 5 and GPT-5.5 alongside open-weight giants like GLM-5.2, a 753B-parameter model you can download and run yourself.
What LLMs can and can’t do
Strong: language tasks of every kind, code generation, and increasingly long agentic work — planning and executing multi-step tasks with tools. Weak: an LLM has no built-in fact checker, so it can hallucinate — state false things fluently and confidently. It only knows what was in its training data (the “knowledge cutoff”) plus whatever you put in its context window, and it predicts rather than calculates, which is why serious deployments pair models with search, code execution and other tools.
See it in practice
Browse the LLM directory for current models with pricing and context specs, the 2026 LLM landscape guide for how the field is structured, or the best LLMs for coding ranking for a concrete buying decision.