Lesson 1.112mIntermediate26.4k studentsFree preview

Tokens and context windows

Models read tokens, not characters, and they can only see a fixed window of them. Both your bill and your failures come back to this.

This lesson sits in LLM Fundamentals, part of Building AI Apps with LLMs. It assumes what came before it and leads directly into the next lesson in the module.

In this lesson you will

  • Understand tokenisation and why cost is per token
  • Budget a prompt against the context window
  • Recognise the failure modes of an overflowing context

Pro tip

Count tokens before you ship a prompt template. Long system prompts are billed on every single call.

Resources