AI model token cost calculator

Language models bill by token, and output tokens cost considerably more than input ones. For one request the amount looks trivial; multiplied by thousands a day it stops being so, and three details can change it for real: prompt caching (repeated context gets far cheaper), reasoning tokens (billed as output even though you do not see them) and batch discounts.

Enter your tokens and per-million prices and you get what each request costs, per day, month and year, how much of the spend is input, output or reasoning, and how the cost would fall if you cached more context. Everything runs in your browser.

Each kind of token is divided by a million and multiplied by its price: uncached input at the input price, input read from cache at its own price, and output and reasoning at the output price. They are added up and, if there is a discount, it is applied to the total. The result is then scaled by the number of requests and days.

The saving from caching is the difference between paying for that share of the input at the normal price or at the cache price.

How a language model is billed

You pay per token, not per request: input tokens (everything you send) and output tokens (what the model generates). Output costs more because generating it takes more compute. A token is a fragment of a word: in English, about 1.3 tokens per word.

Prompt caching

If every request repeats a long block (instructions, examples, a document), many providers let you cache it: later reads cost a fraction of the normal price (around a tenth at several providers). Caching has its rules (minimum size, lifetime, and sometimes a surcharge for writing the cache), but with large, repeated contexts it is the most powerful way to save.

Reasoning tokens

Reasoning models “think” before answering. Those tokens are billed as output even if not shown, and can be far more numerous than the answer. If you use one, estimate how many it generates on average and enter it.

Batch processing

Some APIs cut the price (often by half) for requests that do not need an immediate answer. It suits classifying, summarising or tagging large volumes.

Ways to cut the bill

• Shorten the context you send and avoid repeating long instructions.
• Cache what repeats.
• Use a smaller model for simple tasks.
• Limit the length of the answer.
• Batch whatever is not urgent.

Common mistakes

• Forgetting that output tokens cost several times more than input tokens.
• Ignoring reasoning tokens.
• Calculating with the length of a test prompt rather than the real one (with history and documents).
• Treating prices as fixed: they change.
• Comparing models on the same text without allowing for the fact that they count tokens differently.

Limits of this calculator

• You enter the prices; the calculator neither looks them up nor updates them. The AI cost comparator lists prices per model with their review date.
• It does not include the cache-write surcharge some providers charge, price tiers by prompt length, or tool costs (search, code execution).
• It assumes the same number of tokens in every request.
• It accepts up to 31 days of use per month.

Notice

This is an estimate. Always check prices and terms on the provider’s official page.