What Is Claude's Context Window (And Why Should You Care)?
What Is Claude's Context Window (And Why Should You Care)?
Claude's context window is its working memory for a single conversation. It determines how much information Claude can see and reference at once, including your messages, Claude's responses, uploaded documents, and your Personal Preferences. Claude Opus 4.6 and Sonnet 4.6 have a context window of 1 million tokens (roughly 750,000 words). When the window fills up, the oldest information gets compressed or dropped.
Understanding the context window matters because it affects how sharp Claude's responses are in long conversations, how many documents you can work with at once, and why your Personal Preferences should stay under 500 words. Here's how it works in practice and how to use it wisely.
The Simple Explanation
The context window is Claude's working memory for a single conversation.
Think of it as a desk. Everything in your current conversation sits on that desk: your messages, Claude's responses, any documents you've uploaded, your Personal Preferences, and your Project Instructions. Claude can see and reference anything on the desk.
The desk has a fixed size. When it fills up, the oldest items start falling off the edge. Claude can no longer see them, which means it can no longer reference them.
That's the context window.
How Big Is It?
Claude's context window varies by model:
- Claude Opus 4.6: 1 million tokens (roughly 750,000 words)
- Claude Sonnet 4.6: 1 million tokens
- Claude Haiku 4.5: 200,000 tokens
For comparison, ChatGPT's standard context window is 128,000 tokens. Claude's is roughly 8x larger.
In practical terms, Claude Opus can hold the equivalent of about 10 full-length novels in a single conversation. That's enormous. For most people, you'll never hit the limit in a normal conversation.
But there are situations where it matters.
When the context window matters
Long conversations are the most common case. Go back and forth with Claude for 50+ messages and earlier parts start getting compressed or dropped. Claude uses summarisation to stretch conversations further, but the full detail of early messages doesn't stick around forever.
Large documents eat into it fast too. A single 200-page PDF uses a real chunk of context; upload three and you've burned a lot of working memory before asking a single question. Your Personal Preferences, Project Instructions, and any Skills or custom instructions all load in at the start of every conversation as well, which is exactly why keeping preferences concise matters. A 2,000-word preferences document eats into your context before you've even said hello.
And if you're asking Claude to hold a detailed brief, reference multiple documents, and produce a long output all at once, everything has to fit in the window at the same time.
What happens when you hit the limit?
Claude doesn't crash or stop working. In long conversations it compresses earlier messages into summaries, so you don't lose the general thread, though specific details from much earlier can get fuzzy. Your most recent messages and instructions get priority; Personal Preferences and Project Instructions are always loaded, but very early turns in the conversation are what get compressed first.
You'll usually notice when it happens. If Claude seems to have forgotten something you said 30 messages ago, it's probably been summarised out of the active context. The fix is just to re-state the important bit, or start a new conversation.
How to Use Your Context Window Wisely
You don't need to obsess over token counts. A few habits get you consistently better results.
Keep your Personal Preferences under 500 words, since they load into every conversation, and 2,000 words of preferences is 2,000 words of context gone before you've typed anything. (Our preferences builder aims for the 300 to 450 word range for exactly this reason.)
Build optimised preferences with our free tool →
Start new conversations when you switch topics, rather than running everything through one mega-conversation. Finished discussing marketing strategy and want to debug some code? Start a new chat. Each one gets a fresh window.
For anything ongoing, use a Claude Project instead of re-uploading the same documents every time. Files uploaded to a Project persist across every conversation inside it, so you're not re-loading them and burning context each time.
It also helps to put the important context at the start of your message rather than burying it at the end, and to just be specific about what you need. "Summarise this document in 5 bullet points" uses far less context than "tell me about this document" followed by three rounds of "make it shorter."
The Relationship Between Context Window and Memory
People often confuse the context window with Claude's memory. They're different things.
Context window is temporary. It exists only for the current conversation. When you close the chat and start a new one, the context window resets.
Memory is persistent. Claude remembers things about you across conversations: your name, your role, your projects, your preferences. Memory carries forward even when the context window resets.
Personal Preferences are also persistent. They load into the context window of every new conversation automatically. This is why they're so valuable: they give Claude your context without you having to re-explain yourself, but they do consume a small amount of your context window each time.
The ideal setup is Personal Preferences that are concise but comprehensive (under 500 words), combined with Memory that builds over time, and Projects for topic-specific persistent context. This gives you maximum value from your context window in every conversation.
What this means practically
Three things, if you take nothing else from this. Keep preferences under 500 words, since every word of them loads into every conversation (the free preferences builder is built to get this right automatically). Start fresh conversations when you switch topics instead of letting one run for hours. And for anything spanning days or weeks, use a dedicated Claude Project so the relevant documents persist across every conversation in that workspace instead of getting re-uploaded each time.
You don't need to actively manage the context window. But understanding how it works means you'll naturally use Claude in ways that get better results, and that starts with setting up your preferences properly.