How Claude API pricing works

Every review sends two types of text to the model and receives one back:

• System prompt — your writing guide, voice instructions, program names, and department names. This is sent on every single review and you are charged for it every time. It is currently estimated at ~1,200 tokens but may be higher depending on how much content is in your Settings.
• Your document — the article or page being reviewed. Longer documents cost more.
• The model's response — the review output with all issues listed. This is billed at output token rates, which are 5× more expensive than input tokens.

The three levers you can pull to reduce costs

1. Shorten your system prompt. Go to Settings and audit your writing guide, voice instructions, and lists. Every token you remove saves money on every future review. This is the highest-leverage change because it affects 100% of your calls.
2. Shorten your documents. You don't need to submit an entire webpage for review. If only a section changed, paste just that section. Fewer input tokens = lower cost per review.
3. Control output length. You can add an instruction to your system prompt telling the model to be concise — for example: "Be direct. Flag the problem and give a one-line fix." Shorter outputs save the most money since output tokens are billed at 5× the input rate.

Rough cost reality check: At current Groq free-tier usage levels (~40 reviews/month, average article length), switching to Claude Sonnet would likely cost $3–$8/month. Adjust the sliders above to model your specific situation.