AI Cost Calculators
AI Agent Workflow Cost Calculator
Estimate complete multi-step AI agent costs across planner and worker models, context growth, retries, tool calls, memory, human review, infrastructure, and product margin.
Enter Your Agent Workflow
Model the planner, workers, growing context, retries, paid tools, memory, review, and product economics.
Monthly Runs and Agent Structure
Planner and Worker Tokens
Tools, Memory, and Human Review
Enter current prices for your exact stack
Monetary fields are blank by design. Blank fields are treated as zero, so a partial estimate only includes the rates entered.
Current Model and Service Prices
Product Pricing and Margin
Estimated monthly workflow volume
Successful runs: 9,000
Planner calls: 11,000
Worker calls: 66,000
Paid tool calls: 99,000
Memory operations: 20,000
Human reviews: 500
AI Agent Cost and Unit Economics
The estimate separates monthly operating cost from implementation cost spread across the selected planning period.
Monthly planning cost
Per attempted run
—
Per successful run
—
First-year cost
—
Planner model input
44,000,000 tokens across 11,000 calls
Rate not entered
—
Planner model output
11,000,000 output tokens
Rate not entered
—
Worker model input
181,500,000 tokens including context growth
Rate not entered
—
Worker model output
33,000,000 tokens across 66,000 calls
Rate not entered
—
Paid tool calls
99,000 calls after retry overhead
Rate not entered
—
Memory operations
20,000 reads and writes
Rate not entered
—
Human review
500 reviews · 25 hours
Rate not entered
—
Fixed infrastructure
Hosting, queues, monitoring, databases, or agent platform
Rate not entered
—
Amortized implementation
$0.00 spread across 12 months
Rate not entered
—
Monthly operating cost: —
Operating cost per successful run: —
Break-even price per successful run: —
Price needed for 70% margin: —
Monthly revenue: Add a selling price
Monthly gross profit: Add a selling price
Gross margin: Add a selling price
Price inputs entered: 0 of 9
Budget status: Add a budget to compare
* Important: This calculator stores no built-in model, tool, memory, or labour prices. Enter current official rates for the exact stack being planned. Blank monetary fields are treated as zero, so a partial estimate includes only the prices entered. Average values may not represent expensive long-tail agent runs.
An AI agent is rarely one model request. A single task can include planning, several worker steps, tool calls, growing context, retries, memory operations, and human review. This calculator turns those moving parts into a cost per run, successful task, month, and year.
Planning the Full Cost of an Agent Run
Start with monthly agent runs, steps per run, planner calls, worker calls, and token usage. The calculator models planner and worker models separately because many systems use a more capable model for planning and a lower-cost model for repeated execution.
Add context growth, retry overhead, paid tool calls, memory operations, infrastructure, and human review. The result separates model spend from the wider workflow cost.
One-time implementation cost can be spread across a chosen period to produce a monthly planning cost instead of hiding setup work outside the estimate.
Modelling Context Growth Across Steps
Later agent steps often receive more context than the first step. Tool responses, retrieved documents, intermediate plans, and earlier outputs can all be carried forward.
Enter the base worker input tokens and the extra input tokens added at each step. The calculator uses an increasing sequence rather than assuming every worker call has the same input size.
This is especially useful for research agents, coding agents, browser agents, support workflows, and multi-tool automation.
Including Tools, Memory, and Human Review
Paid search, browsing, data enrichment, code execution, maps, email, documents, databases, and other APIs can add a separate charge to each run.
Memory may also create vector reads, writes, database operations, or managed-platform charges. Enter these as a price per 1,000 operations.
Human review is calculated from the share of runs reviewed, average review time, and hourly labour rate.
Using the Unit Economics Result
The calculator shows operating cost per attempted run and per successful run. Add the amount charged for each successful task to estimate monthly revenue, gross profit, and gross margin.
The break-even price covers the entered planning cost. The target-margin price adds the headroom needed for the selected gross-margin goal.
These figures help decide whether to reduce steps, move worker tasks to a lower-cost model, limit retries, shorten context, replace paid tools, or change product pricing.
Practical Decisions This Tool Supports
- Estimate the cost of a multi-step agent before launch.
- Separate planner and worker model economics.
- Measure the effect of context growth and retry loops.
- Add paid tools, memory, infrastructure, and human review.
- Calculate cost per successful task rather than only per run.
- Find the price needed for a target gross margin.
- Check monthly budget and first-year spend.
Costs and Risks Outside the Estimate
The result does not automatically include payment fees, taxes, customer support, sales, refunds, free-tier abuse, observability, data transfer, security review, compliance, downtime, or the business cost of incorrect agent actions.
Average values can hide expensive long-tail runs. Review median, high-percentile, and worst-case production data when setting limits and pricing.
Frequently Asked Questions
Why is an AI agent more expensive than a single model call?
An agent can make several planner and worker calls, carry growing context across steps, call paid tools, retry failed actions, use memory, and send some runs to human review. The calculator includes these layers separately.
What does context growth per step mean?
Tool results, prior decisions, retrieved data, and intermediate outputs can make later model calls larger than earlier calls. The calculator adds the entered number of input tokens to each later worker step.
How should I estimate retry overhead?
Use logs when available. Include repeated model calls and tool calls caused by timeouts, invalid output, failed actions, self-correction, or agent loops. A 10% value means the model and tool workload is increased by 10%.
Why are provider prices blank?
Agent workflows can use different models and paid tools. Blank rates prevent example prices from appearing as verified live prices. Enter current official rates for the exact planner, worker, tool, memory, and labour services being considered.
What is the difference between cost per run and cost per successful run?
Cost per run divides spend across every attempted run. Cost per successful run divides the same spend only across the runs expected to complete successfully, so it reflects failure and abandonment.
How is the price needed for target margin calculated?
The calculator divides planning cost per successful run by one minus the target gross-margin percentage. This is a planning price before payment fees, tax, support, sales, and other business costs.