AI & LLM cost
Per-request cost
The full cost of one API call, shown as one number. It includes input tokens, output tokens, and any tool calls.
Last updated
Definition
Per-request cost turns token-level pricing into a number that a product or engineering decision can use. "This feature costs $0.004 per completion" is something you can act on. "$3 per million output tokens" is not, on its own. Per-request cost is the AI and LLM equivalent of unit economics. It works best when you track it per feature or per workload, not as one blended average.
Set a rough expected per-request cost before a feature ships. Then a later regression shows up at once. A prompt may have grown without anyone noticing. A model swap may not have been checked for cost. Without the expected number, you find out a month later on the invoice.
Where it shows up
Per-request cost is a computed figure. Take the input and output token counts of one request. Multiply them by the provider's published per-token rates. Add the cost of any tool calls it made. The usage object in the API response already gives you the token counts for that one call.
What makes it expensive
Teams track one blended average across a whole product instead of a figure for each feature. That average hides one feature's cost regression inside a total that looks stable. Write down what a feature should cost per request before it ships. Without that number, a regression has nothing to fail against.
Related
Token spend moves every day.
Model prices and rate limits change between invoices. CostMon tracks your AI spend daily, so a change shows up before the bill does.