A Large Language Model is a neural network, typically based on the transformer architecture, trained on large corpora of text to predict the next token given a context. The scale of both the training corpus (typically trillions of tokens) and the model itself (typically tens of billions to trillions of parameters) gives LLMs emergent capabilities not present in smaller models: reasoning, code generation, summarization, translation, and sustained multi-turn conversation.
The practical frontier as of 2026 includes Anthropic's Claude family, OpenAI's GPT family, Google's Gemini family, and Meta's Llama family, along with Chinese competitors including DeepSeek, ByteDance's Dola-Seed, and Z.ai's GLM. Commercial LLM deployment is the foundation of most enterprise AI products, from copilots to autonomous agents.
Modern LLMs are typically multimodal — accepting and generating not just text but images, audio, and structured data — which has made the term "language model" partially a misnomer. The name has stuck because the architectural lineage traces back to text-only predecessors.