AI API cost calculator

Know the bill before you build

Large language model APIs charge by the token, a small piece of text of roughly three to four characters in English. The cost of one request is tiny, a fraction of a cent, which makes it easy to ignore until a product goes live and thousands of requests turn into a real monthly bill. Working out the cost early helps you choose a model, decide how long your prompts can be and set a price for your own product that leaves a margin.

How the cost is calculated

Providers publish two prices per model, usually per million tokens: one for input, the text you send, and a higher one for output, the text the model writes. The cost of a request is the input tokens times the input price plus the output tokens times the output price, divided by a million. Multiply by requests per day and the days in a month and you have the monthly figure. A discount field covers batch processing or cached input, when your provider offers them.

The prices are yours to enter

This calculator does not ship with a price list, on purpose. Prices change often, differ between models and tiers, and may include extras such as cached input, long-context surcharges or regional pricing. The numbers pre-filled in the form are only examples so that you can see how it works. Copy the current rates from your provider's pricing page for a real estimate.

Reading the result

• Cost per request, day, month and year for your usage.
• Tokens per month, split into input and output, useful for checking quotas and rate limits.
• Input and output share, which shows what drives the cost. Long answers cost more than long questions because output tokens are pricier.
• Comparison. Fill in a second model to see the monthly difference and which one is cheaper for your usage.

Tips to lower the cost

Shorter prompts and instructions, trimming the context you resend, limiting the length of answers, caching repeated input and using a smaller model for simple tasks are the usual levers. If you do not know how many tokens your text uses, the token counter gives an estimate.

Limits

It is an estimate. Real bills depend on the exact tokenizer, on tokens you do not see, such as system messages and tool definitions, on retries, and on features billed separately such as images, audio or tool calls. Nothing you enter leaves your browser.