Free tool · by Zenrows
Count the tokens in any prompt.
Paste two versions, a verbose prompt and a tight one, or raw HTML and clean markdown, and see how much context budget you save. Counts update as you type.
Runs entirely in your browser. Token counts are estimates (≈ 4 characters / token); exact counts vary by model tokenizer. Nothing you paste leaves this page.
Every token is context you pay for.
Prompts and pages compete for the same window. Trim the noise and you fit more signal, spend less, and answer faster.
Budget
Fewer tokens in means a smaller bill on every single call.
Window
What you trim is room the model gets back for real content.
Speed
Less to read means lower latency on every request.
Where the savings come from.
Most token waste is structure, not meaning. Cut it before the model ever reads it.
01
Drop the markup
Raw HTML is mostly tags. Clean markdown says the same in a fraction of the tokens.
02
Cut the boilerplate
Nav, cookie walls and footers carry no signal. Strip them before they cost you.
03
Tighten the prompt
Say it once, plainly. Shorter instructions often read clearer to the model too.
Get clean, small payloads by default.
Fetch returns pages already stripped to markdown, so the token bill starts low before you write a single prompt.
# clean markdown back, ready for the model curl -G "https://api.zenrows.com/v1/" \ -d "apikey=ZR_••••••••" \ -d "url=https://example.com/page" \ -d "response_type=markdown"