How context windows are measured
Context windows are measured in tokens. A token is roughly three-quarters of a word in English. So 1,000 tokens is approximately 750 words, or about 3 to 4 pages of text. When a model has a "128K context window," it means it can process about 128,000 tokens (roughly 96,000 words) in a single conversation.
The context window includes everything: the system prompt, your uploaded files, your messages, and the AI's responses. As the conversation grows, the available space for new content shrinks.
Context windows by model
Different AI models have different context window sizes. Claude, made by Anthropic, offers a 200,000-token context window, which holds roughly 150,000 words. ChatGPT with GPT-4o has a 128,000-token window, good for about 96,000 words. Google's Gemini models offer context windows ranging from 128,000 to over 1,000,000 tokens depending on the version.
Larger context windows let you include more reference material in a single session. This is particularly valuable for tasks like literature reviews, long document analysis, or working with an entire codebase.
What happens when you exceed the context window
When the total content in a conversation exceeds the model's context window, the AI begins to lose access to earlier parts of the conversation. In practice, this means the AI might "forget" instructions you gave at the beginning, miss details from earlier uploaded files, or contradict information it provided in earlier messages.
This is not a bug. It is a fundamental constraint of how current AI models process text. They can only attend to a fixed amount of content at a time.
Practical tip: If your project requires more context than fits in one window, prioritize the most relevant files and instructions. A well-curated 50,000-word bundle often produces better results than a 150,000-word dump that includes irrelevant material, because the AI can focus its attention on what matters.
Why context windows matter for your workflow
Understanding context windows changes how you work with AI. Instead of uploading everything and hoping the AI figures it out, you learn to be deliberate about what goes into the window. You curate your context the way you would prepare a briefing document for a colleague: include the essentials, leave out the noise, and organize it so the reader can find what they need.
This is the core idea behind context engineering: the practice of deliberately deciding what goes into the AI's context window, in what format, and in what order.
How to check your token count
Most AI platforms do not show your token usage in real time. If you are preparing a large context bundle, it helps to check the size before uploading. Tools like Riflet show a live token estimate as you select files, so you know whether your bundle fits within your model's context window before you start the conversation.
As a rough guide: a 10-page document is about 4,000 tokens. A 50,000-word manuscript is about 67,000 tokens. A folder of 20 research notes averaging 2 pages each is about 16,000 tokens. These are well within the limits of both Claude and ChatGPT.
Make the most of your context window
Riflet helps you build token-aware context bundles that fit perfectly within any AI's limits.
Download Riflet free