TokenPad

Models

Large language model (LLM)

A large language model is a neural network trained to predict the next token in a sequence, which turns out to be enough to produce useful text, code and reasoning.

Also written: LLM

Everything an LLM does is next-token prediction. Answering a question, writing code and following instructions are all the same operation applied repeatedly, with each generated token becoming part of the input for the next.

That single mechanism explains most of the behaviour people find surprising: why output length drives latency almost linearly, why asking a model to show its working improves accuracy, and why it will confidently produce a plausible next token even when no correct one exists.

In practice

The commercially relevant fact is that capability and price no longer move together in a simple way. Classification, extraction, routing and formatting are close to saturated on small models, and the price gap between tiers is often more than tenfold. Sending easy traffic to a frontier model is the single most common source of an avoidable bill.

Common questions

Do I need the most capable model?

Usually not. Classification, extraction, routing and formatting are close to saturated on small models, and the price gap between tiers is often more than tenfold. Reserve the frontier tier for open-ended reasoning.

How do I choose between models?

On price, context window and tokenizer availability, which are facts — then on your own evaluation set, which is the only measurement of quality that applies to your data. Leaderboards are not a prediction about your task.

Measure it

  • TokenA token is an entry in the fixed vocabulary a language model was trained with — the unit it reads, and the unit you are billed in.
  • TransformerThe transformer is the neural network architecture behind essentially every current language model, built around a mechanism called attention.
  • InferenceInference is running a trained model to produce output, as opposed to training, which is the process that produced the weights.
  • Context windowThe context window is the total token budget for a single request, covering the system prompt, conversation history, tool definitions, documents and the response the model generates.

More in Models