AI API COST CALCULATOR
A headline price is not a workload budget. Enter the token mix your application actually produces, choose a listed model, and see the estimated USD cost for one request and a monthly request volume.
Create an accountAI APIs normally charge input and output at different rates. The basic estimate is (input tokens ÷ 1,000,000 × input rate) + (output tokens ÷ 1,000,000 × output rate). Keeping the two sides separate prevents a short prompt with a long answer from looking artificially cheap.
Measure a small sample of successful requests instead of guessing from characters or message count. Include system instructions, conversation history, retrieved context, tool results, and the assistant output when estimating production traffic.
Prompt caching, retries, and provider-specific billing can change the final total. Use the calculator as a transparent baseline, then reconcile it against the request-level usage ledger and your application's retry behavior.
Multiply the blended request estimate by expected successful requests. Add a separate allowance for retries and long-tail workloads, then use project keys, balance alerts, and hard caps to keep the real spend close to the budget.
INTERACTIVE ESTIMATE
Choose a listed model, enter the tokens your app sends and receives, and adjust the monthly volume. The estimate stays in USD and keeps input and output visible.
Formula(input ÷ 1M × input rate) + (output ÷ 1M × output rate)
THE BASELINE FORMULA
Start with successful requests, then add an explicit retry and long-context allowance. Prices are USD per one million tokens.
input_cost = input_tokens / 1_000_000 * input_rate
output_cost = output_tokens / 1_000_000 * output_rate
request_cost = input_cost + output_cost
monthly_cost = request_cost * successful_requestsCOMMON QUESTIONS
Estimated cost = (input tokens ÷ 1,000,000 × input price) + (output tokens ÷ 1,000,000 × output price). Add cache, retries, and other applicable charges separately when the provider bills them.
Usually not. Output tokens are often priced higher, so calculate both sides separately using the current model rate.
No. It is a planning estimate in USD. The applicable published model rate, usage record, exchange rate, payment method, and billing rules determine the final account charge.
Measure average input and output tokens per successful request, multiply the blended cost by expected requests, and add a conservative allowance for retries and unusually long jobs.
Use the live WorldGate pricing page and the individual GPT or Claude model pages before funding or comparing a production workload.
Lower model cost. Clearer usage.