Start with the usage record
Your API response or provider dashboard should report input and output usage separately. Input includes instructions, the current message and any conversation history sent again. Output includes generated text and, for some models, reasoning tokens. A total token count hides the split, so a precise cost calculation starts with both counts.
Use the formula
USD = (input tokens × input price per million + output tokens × output price per million) ÷ 1,000,000. For example, assume 2,000 input tokens at $3 per million and 500 output tokens at $15 per million. Input costs $0.006; output costs $0.0075; the request costs $0.0135. These are illustrative rates, not a quote for a named model.
Scale a request into a budget
At 10,000 identical requests, the example costs $135 before discounts or other billable services. Use real usage samples, not only a best-case prompt. Conversation history can grow on later turns, and retries increase the number of requests. For a monthly forecast, multiply each observed token class by expected monthly request count before applying its rate.
Check billing exceptions
Provider price pages may list separate rates for cached input, batch processing, long context, images, audio, tool calls and service tiers. A text-token estimate omits those categories. Check the exact model ID, route and billing date on the provider's price page before treating an estimate as an invoice. Open the Tokens to USD calculator for current catalog rates and separate input/output entry.