Get started
Context windows
5 min read
Every AI has a memory limit. Understanding it changes how you use every tool.
The bucket
Context window
1M tokensSystem prompt~6K
Conversation history~38K
Pasted document~51K
Room for response~33K
danger zone
When the bucket overflows, oldest messages get silently dropped.
How models compare
1 token ≈ ¾ of a word
GPT-5.4
1M
~750K wordsClaude Opus 4.6
1M
~750K wordsGGemini 2.0
2M
~1.5M wordsWhat actually happens
Short conversationAI remembers everything
Long document pasteLess room for back-and-forth
Marathon sessionAI forgets early context, may contradict itself
Work with the limit
Start fresh for new topicsNew chat = full context window
Summarize, don't pasteAsk AI to summarize long docs first
Front-load important infoPut key details at the start, not the end