Skip to content
S/T
Free prompt audit

Your system prompt is quietly running up your AI bill. See exactly how much.

Paste your prompt. Get a ranked list of what’s wasting tokens, each priced in dollars per month. Plus a cleaner version you can copy, and what it would cost on other models.

No account, no upload. Your prompt never leaves this browser.

01 / Paste

stays in your browser
How often this prompt is sent.
Average completion length.
Leave at 0 if you are not caching yet.
The tokenizer loads when you paste something, not before.

Nothing to audit yet

Paste a system prompt above, tell it how often you send it, and you will get a priced list of what is wasting money — plus a rewritten version you can copy. No account, no upload, no model call.

Examples are illustrative prompts written for this tool, not real customer prompts. Prefer not to paste at all? Run the same analysis locally with the CLI.

What it looks for

26 patterns of waste, from small stuff like filler words and smart quotes to the expensive ones: content that can’t be cached, tool definitions nobody has priced, and output rules written in prose that a schema would handle for free.

Browse all 26 patterns

Why the numbers differ per model

The same text isn’t the same number of tokens on every model. Newer Claude models count about 30% more than older ones, so switching models can raise your bill without changing a word. We re-count for each model instead of reusing one number.

Why models count differently

Where the prices come from

38 models from Anthropic, OpenAI and Google, priced by hand from their published pages and dated. Last checked 2026-08-10. Token counts are exact for OpenAI and clearly labelled estimates for the rest.

Read the methodology