Help - AI Agent Token Cost Estimator (token_cost)

Click here to open the calculator: AI Agent Token Cost Estimator

AI Agent Token Cost Estimator

Purpose

token_cost() estimates the input, output, and total API cost of an AI agent over multiple calls or rounds.

It supports two common request patterns:

  • Full history resent: each call includes the growing conversation history.
  • Only current input: each call includes the initial context and only that round's new input.

The estimate assumes the same number of new input and output tokens in every round.


Inputs

  • Number of rounds: The number of API calls.

  • Initial context tokens: Tokens sent as fixed context on every call, such as the system prompt, instructions, tool definitions, or other unchanged context.

  • New input tokens per round: New user, tool, or other input tokens added during each round.

  • Output tokens per round: Assistant output tokens generated during each round.

  • Input price per million: The provider's price for one million input tokens.

  • Output price per million: The provider's price for one million output tokens.

  • Input mode: a) Full history resent (default) - each request includes the initial context, all inputs through the current round, and all earlier assistant outputs. b) Only current input - each request includes the initial context and only the current round's new input.


Results

The calculator returns:

  • Cumulative Input Tokens
  • Cumulative Output Tokens
  • Input Cost
  • Output Cost
  • Total Estimated Cost

Costs are returned in USD.


How the Estimate Is Calculated

Let:

  • n = number of rounds
  • S = initial context tokens
  • u = new input tokens per round
  • v = output tokens per round
  • P_i = input price per million tokens
  • P_o = output price per million tokens

Only current input

Full history resent

At round k, the request contains:

Therefore:

For either mode:

The current-input estimate grows linearly with the number of rounds. The full-history estimate has a quadratic input component because earlier messages are billed again when resent.


Example

With the defaults, full mode estimates 227,500 input tokens, 10,000 output tokens, and a total cost of 1.11 USD. Using current mode reduces the input total to 85,000 tokens and the estimated cost to 0.54 USD.


Notes and Limitations

  • The calculator uses fixed token counts per round; actual conversations usually vary by call.
  • Prompt caching, cached-input discounts, batch pricing, tool charges, and other provider-specific pricing are not included.
  • In full-history mode, all resent input is priced at the standard input rate, so the result is an uncached baseline.
  • Use token counts reported by the API for the most accurate final cost.