Claude guide

Claude context window limit — what it means and how to handle it

Claude has one of the largest context windows of any AI assistant — but it still has a limit, and hitting it creates real problems for anyone doing deep, extended work. Here's what the limit means, when you'll feel it, and what to do about it.

Start free on Chrome

Claude's context window: 200,000 tokens

  • Claude's context window is 200,000 tokens — roughly 150,000 words, or about 500 pages of text. For the vast majority of Claude users, you'll never hit the hard limit in normal use.
  • But there's an important caveat: even well before the 200K limit, performance can degrade in very long conversations. This 'context rot' means Claude pays more attention to recent content and can lose track of instructions or facts given much earlier.

When you'll actually feel the limit

  • Very long documents: uploading large PDFs or codebases can consume a significant portion of the window before the conversation starts.
  • Extended multi-session work: accumulating many messages and large outputs over time.
  • Context rot in long sessions: subtle degradation where Claude ignores early instructions, even before the hard limit.

How to handle it

  • Use Claude Projects for ongoing work — instructions and files stored separately from the active context.
  • Enable Code Execution in Settings → Tools for automatic context management on paid plans.
  • When approaching the limit, ask Claude to summarise everything and restart in a new conversation.
  • Use thredly for automatic session capture — so your context is always ready when you need to start fresh.

Related guides

Frequently asked questions

Does the context window reset between conversations?

Yes — each new conversation starts with a completely fresh 200K window. Previous conversations don't carry over unless you manually paste in context, or use Projects.

What's the difference between Claude's context limit and Claude Projects?

Projects store files and instructions separately, using retrieval to load relevant content rather than dumping everything into the active context at once. This means Projects can handle much larger total amounts of information than the raw context limit suggests.