LLM (Large Language Model)
A neural network trained on text that predicts the next token, used as the engine inside agents.
Last updated: April 26, 2026
Definition
An LLM is a transformer model trained on billions to trillions of tokens that predicts the next token given the previous ones. The model is the same primitive whether it is generating text, calling tools, or writing code. It just predicts tokens. What changes between use cases is the prompt and the surrounding code. Modern frontier models (Claude Opus 4.6, GPT-5.4, Gemini 2.5 Pro) all have ~200K-2M context windows, native tool use, and multimodal input. The differences between them at the frontier are real but smaller than the differences between any frontier model and a 7B open-source model.
When To Use
The default reasoning engine for any AI feature. Pick model size by task: frontier for hard reasoning, mid-tier for most production work, small for classification and routing.
Building with LLM (Large Language Model)?
I've shipped this pattern in real production systems. If you want a second pair of eyes on your architecture, that's what I do.