Does the context window include the model's answer?

Yes. Input and output share the limit, so a prompt that fills the window leaves no room for the reply.

Is a bigger context window always better?

No. You pay for every token you send, and models use information in the middle of very long prompts less reliably. Send the part of the page that matters.

Glossary

Context window

The model's working memory, measured in tokens and paid for by the token.

By Roger Campos · Last updated: September 2026

TL;DR

A context window is the maximum amount of text, measured in tokens, that a language model can consider in one request — the instructions, the documents you include, the conversation so far and the model's reply together. Anything past it is cut off or refused, and everything inside it is billed.

Start free See the API

Free plan, no credit card. 1,000 credits a month.

How it works

Tokens and limits

Models read tokens, not characters: pieces of words from a fixed vocabulary. For English text a token averages roughly four characters, so 1,000 tokens is about 750 words; code, URLs and non-Latin scripts use more. A model's context window is a hard limit on input plus output, and providers bill per token, so a bigger window is not a reason to fill it.

Long contexts also read worse. Models are measurably better at using information near the start and end of a long prompt than in the middle, so a page's key paragraph buried in 50,000 tokens of markup is easier to miss.

In practice

Where web pages go wrong

Raw HTML is the expensive way to give a model a page. Markup, inline scripts and styles dwarf the text: measured over pages URLpipe processes, a page's readable text is about 6% of its raw HTML. Send the HTML and you pay for the other 94% and dilute the part that matters.

Convert to Markdown and keep only the main content.
Remove what you know is irrelevant — cookie policies, related posts — before converting.
When the page is still too long, chunk it and retrieve only the relevant parts.
When you only need the gist, summarize once and reuse the summary.

URLpipe

With URLpipe

/markdown is the compact input: the rendered page's main content, at 1 credit. /summarize returns a short summary instead, for when that is the thing you want to produce. Both work on pages that need JavaScript. For agents, the MCP server exposes the same as fetch_markdown and summarize_page.

Try it

Context window, on a page you choose

Turn any URL into clean Markdown

Paste a link and get the page's main content as tidy Markdown — headings, lists, links and code kept, navigation and cookie banners stripped. The format LLMs and RAG pipelines work best with.

Try it free

Related terms

Markdown for LLMs Chunking Embeddings

FAQ

Frequently asked questions

It depends heavily on the page and the format. The same article can be tens of thousands of tokens as raw HTML and a few thousand as Markdown of its main content.

Keep reading

Chunking web pages for RAG Request caching with max_age HTML to Markdown for LLMs LlamaIndex A sign-in form or an error page is a failure, not a result Keyword extraction All of glossary

See it on your own pages.

Free plan, no card. Confirm your email and your API key is live — you'll be making real requests in minutes.

Start free Email the team