Context window

What counts toward the context window?

According to the Claude documentation, everything in a request takes up space. That includes the system prompt, every message in the conversation, tool results, images and documents, and the tool definitions. The text the model generates counts too, including its reasoning. In a long conversation the window fills up turn by turn.

Per Anthropic’s models overview (as of October 2026), the current Claude models have a 1 million token window, and Claude Haiku 4.5 has 200,000. For the current tokenizer, 1 million tokens equal roughly 555,000 words. Other vendors use other sizes, and the numbers change often.

What happens when the limit is reached?

If the input alone is larger than the window, the Claude API returns a 400 error (invalid_request_error, “prompt is too long”). If the input fits but generation reaches the limit, newer models stop with the stop reason model_context_window_exceeded. In chat apps, older parts of a conversation may be dropped or summarized, depending on the product.

A bigger window does not automatically give better results. Anthropic says accuracy and recall degrade as the token count grows and calls this effect context rot.

What does it mean for SEO?

AI search systems and agents load page content into the window. Because space is limited, many systems work with selected passages instead of whole pages (assessment). Compact, clearly structured pages are easier to use in that setting.

Example

An agent compares three product pages and a price list. The system prompt, the question, the four documents and the answer all share one window. If one page is very long, less room remains for the others.

FAQ

What is the context window of ChatGPT? It depends on the model and the product. Check the vendor’s current model documentation.

What happens when the context window is full? Through an API, an oversized input causes an error. In chat apps, older messages are dropped or summarized.

Last updated: October 7, 2026