How to use it
- Pick the model you plan to use. Its prices fill in, and you can edit them if you have a special rate.
- Enter the average input and output tokens per request and how many requests you expect per day.
- If part of each prompt repeats, enter the share that can be cached.
- Read the cost per request and per month, and check the list of alternatives below.
How the cost is calculated
Each request has an input cost and an output cost. Input cost is the number of input tokens multiplied by the input price, divided by one million. Output cost is worked out the same way with the output price. If some of the input is served from the provider's prompt cache, that share is charged at the cached rate instead. The calculator adds these up for one request, then multiplies by your daily volume. A month is taken as 30.4 days, the average month length.
Worked example
An internal assistant for a 40-person sales team gets about 600 questions a day. Each request sends 3,500 tokens: a 3,000-token product guide and a 500-token question. Each answer averages 350 tokens. On a model priced at $3 input and $15 output, one request costs $0.0105 + $0.00525 = about $0.0158, or roughly $287 a month. Because the product guide is identical every time, it can be cached. With 85 percent of input cached at $0.30, the cost per request falls to about $0.0077, and the monthly bill to roughly $141.
Getting realistic averages
The quality of any estimate depends on the token numbers you put in. Paste a real example request into the token counter to measure input. For output, look at a few real answers rather than guessing. Remember that chat apps resend earlier messages, so the average input grows as conversations get longer.
When the estimate will be off
- Reasoning models bill hidden thinking tokens as output, which can multiply the output cost.
- Retries and errors are billed too. Add a few percent if your app retries failed calls.
- Tool use such as web search may carry a separate per-call fee.
- Special pricing such as batch discounts, regional pricing or enterprise agreements is not included unless you edit the prices.
Choosing a cheaper model
The list under the calculator ranks other models by what your exact workload would cost on each. Price is only half the decision: a cheaper model that needs two attempts or longer prompts to get the right answer can end up costing more. Test the top two or three candidates on a sample of real requests before switching.
Questions
Does this include taxes or currency conversion?
No. Results are in US dollars before tax. If you pay in another currency, multiply by your card or bank exchange rate.
What is prompt caching?
When the beginning of your prompt is identical to a recent request, many providers reuse their earlier work and charge a much lower rate for that cached part. It suits apps that send the same long instructions or documents every time.
Why is my real bill higher than the estimate?
The most common reasons are longer outputs than expected, chat history growing over a conversation, reasoning tokens, retries, and separately billed tools. Compare your provider's usage report with your averages and adjust.
Can I enter my own negotiated prices?
Yes. After choosing a model, type over any of the prices. Use Reset prices to go back to the listed rates.