- Set the workload
Enter agent runs and any other volume assumptions that define the period or batch.
- Add usage assumptions
Complete the remaining usage fields, keeping token, duration, or size units consistent with their labels.
- Enter current rates
Use the current prices for the exact model, endpoint, quality, and processing tier you plan to use.
- Review the estimate
Read the main result and the supporting breakdown to see the largest cost or capacity driver.
- Test another scenario
Change one assumption at a time or select Reset to restore the default example values.
OpenAI Agent Calculator
The OpenAI Agent Calculator estimates the cost of a multi-step agent run. It accounts for planning steps, tool calls, model calls per step, token usage, per-call tool fees, and run volume.
Agent workflows often cost more than single-turn chat because one user request can trigger loops, retries, searches, code execution, or delegated tasks. This calculator makes those multipliers explicit and helps teams test whether reducing steps or tool frequency materially changes operating cost.
Calculation inputs
runs
steps
calls
tokens
tokens
USD
USD
calls
USD
Result
—
Estimated result
Model calls—
Paid tool calls—
Model cost—
Cost per agent run—
Model calls = Runs × Steps × Model calls per step. Tool calls = Runs × Steps × Paid tool calls per step. Total cost = Model-call token cost + Tool calls × tool fee.
The calculator applies the entered values directly and displays rounded results for planning.
What the result means
The result estimates model and paid-tool charges across all agent runs.
Infrastructure, vector databases, third-party APIs, human review, and failed-run remediation are not included unless represented as a tool fee.
Given: Example assumptions
Calculation: For 10,000 runs averaging five steps, 1.2 model calls and 0.4 paid tool calls per step, the workflow makes 60,000 model calls and 20,000 tool calls. With 1,800 input and 450 output tokens at $2.50 and $15 per million, model cost is $675. At $0.0025 per tool call, tool cost is $50, for a $725 total.
Result: The displayed result follows the same formula and rounding logic.
What counts as an agent step?
Use one reasoning or action stage in the workflow. A step can contain more than one model or tool call.
Should retries be included?
Yes. Increase average steps or calls per step to reflect retries, validation loops, and recovery behavior.
How do I model tools with different prices?
Calculate each tool category separately or use a weighted average fee based on expected call mix.
Does this include long-running background infrastructure?
No. Add hosting, queues, observability, databases, and worker costs separately.
Which metric is most useful for optimization?
Cost per successful run is usually more actionable than cost per call, especially when workflows have variable completion rates.