Context Window

Context Window is the maximum amount of text an AI model can process in a single conversation, measured in tokens. Tokens don't perfectly map to words — in English, roughly 3/4 of a word per token; in Chinese, about half a character per token.

Claude's Context Window is 200,000 tokens — one of the largest among mainstream models. Comparison: GPT-4o is 128K tokens; Gemini 1.5 Pro standard is 128K (Pro can reach 1M but slower).

What does Context Window include? Everything you've typed in this conversation, all of Claude's replies, any uploaded document content, Claude Projects Instructions and knowledge base files — all count toward Context Window usage.

What happens when you exceed the limit? Context Window is a rolling window. When total content exceeds 200K tokens, the earliest conversation content gets pushed out of view — Claude doesn't refuse to answer; that content is just outside its visual range. You may notice it "forgetting" things you said at the very start of the conversation. This is the typical sign of a full Context Window.

Practical sense of scale: 200K tokens fits a 400-page book, the key files of a small codebase, or 8 continuous hours of verbatim transcript. For most everyday tasks, you'll rarely hit the Context Window ceiling.

What's the difference between Context Window and the memory feature?

This is a concept many people confuse. Context Window and memory are two completely different things:

Context Window: everything in this conversation. When the conversation ends, Context Window content disappears — start a new conversation and Claude has no idea what you previously said. Every new conversation is a completely fresh start.

Memory feature: personal information Claude extracts and saves from past conversations for use across conversations. Claude knowing your name, work background, and preferences comes from the memory system — not because that information is in the current Context Window.

Claude Projects Instructions: similar to "permanent system settings" — every time you start a conversation in the same Project, Instructions automatically load into the Context Window. This saves you from re-explaining context each time, but these Instructions do consume Context Window space.

A memory analogy: Context Window is all the documents you've spread out in this meeting (packed away when it ends); memory feature is important notes in your notebook (still there next time); Projects Instructions are the standing meeting materials you always bring.

Why does Claude sometimes "forget" things said earlier in very long conversations?

This is Context Window saturation in action. When accumulated conversation tokens approach or exceed the 200K limit, Claude can no longer see the earliest conversation content — it's not selectively forgetting; that content is simply outside its processing range.

Several practical solutions:

Strategy 1: Start new conversations proactively. Break large tasks into multiple conversations, each focused on one subtask. Don't try to do everything in one ultra-long conversation.

Strategy 2: Make periodic summaries. When a conversation gets long, have Claude generate a "current key points summary," then start a new conversation with the summary as background context.

Strategy 3: Use Claude Projects for persistent context. Put background information you want available in every conversation into Project Instructions — regardless of how many new conversations you start, Claude can access this information without re-explaining each time.

Check Context usage in claude.ai: You can view current Context Window usage percentage in conversation settings, helping you predict when to start a new conversation or make a summary.

How much does Context Window size matter when choosing AI tools?

For regular users, Context Window size differences don't matter much for most everyday tasks — work emails, daily Q&A, short-form writing are nowhere near 128K, let alone 200K.

Scenarios where Context Window genuinely matters: analyzing long documents — contract review, research reports, codebase analysis. If you need AI to read a 200-page contract at once, a 128K model may need to process in segments, while Claude's 200K handles it in one go. Long conversation accumulation — tasks requiring many conversation rounds to complete (like gradually designing a complex system) benefit from larger Context Windows maintaining consistent understanding throughout. Multi-document integrated analysis — analyzing multiple documents simultaneously for connections; larger Context Windows let you load more material at once.

Practical recommendation: if your work primarily involves these scenarios, Claude's 200K is a meaningful advantage. If your work is mainly daily Q&A and short-form tasks, Context Window size doesn't affect you much — make decisions based on other factors like output quality and cost.