Tokens & the context window — Foundations
AI reads in tokens, not words
Models chop text into tokens — chunks that are often a whole word, sometimes part of one. On average a token is about ¾ of a word, so 1,000 words ≈ 1,300 tokens.
Tokens are the unit for everything: how much the model can read, how fast it writes, and what the API bill says.
📝 Read the full lesson notes
AI reads in tokens, not words
Models chop text into tokens — chunks that are often a whole word, sometimes part of one. On average a token is about ¾ of a word, so 1,000 words ≈ 1,300 tokens.
Tokens are the unit for everything: how much the model can read, how fast it writes, and what the API bill says.
The context window is its working memory
The context window is how many tokens the model can consider at once — your messages, its replies, pasted documents, uploaded files and hidden instructions all included.
Today's models hold roughly 100,000 to over a million tokens — hundreds to thousands of pages. Big, but not infinite.
Why long chats get worse
In a very long conversation, old details get compressed or pushed out, and the model has more noise to sift through. That's when it starts forgetting instructions, repeating itself or mixing up topics.
Pro move: one chat per task. When you switch topics — or things get weird — start a fresh chat and paste in just what matters.
Context habits that instantly help
- New task → new chat. It's free and it resets the clutter.
- Paste the relevant part of a document, not the whole thing, when you can.
- Put long material first and your question at the end.
- Use Projects (ChatGPT, Claude) or Gems (Gemini) for material you reuse, instead of pasting it every time.
Part of 🧠 Foundations in the free AI Bootcamp.