Context Windows Explained: How Much Can an AI Read at Once?

A model’s context window is everything it can “see” while answering: its instructions, your messages, its own earlier replies and any documents you attached — all counted in tokens.

How big?

Early chat models handled a few thousand tokens — a few pages. Many current models handle hundreds of thousands of tokens, and some around a million: whole books or large codebases. Check your tool’s documentation; free tiers often allow less than the model’s maximum.

What happens at the limit

• Some tools refuse to send more.
• Many chat apps quietly drop or summarise the oldest messages — the AI “forgets” what you said at the start.
• Answers may cover only part of a long document.

Bigger isn’t automatically better

Models can miss details buried in the middle of very long inputs, and every extra token costs time and money. A focused 5-page excerpt often beats a 500-page dump.

Practical habits

1. Start a new chat per task. Old, unrelated messages still take up space and can confuse the answer.
2. Put key instructions last as well as first in very long prompts.
3. Summarise as you go: ask for a summary of decisions so far and paste it into a fresh chat.
4. Split long documents by section and ask the same question of each, then combine.
5. Use retrieval for large document sets — search first, send only the relevant passages (RAG course).