AISynq
AI unit economics

How much does an AI agent cost per task?

An agent that plans, calls tools across several turns, and completes a task end to end.

The short answer

Per call$0.108
Per user, per month$3.24
Of a $199.00 price1.6%

On Claude Sonnet 5, at 30,000 input and 1,200 output tokens per call. Thirty thousand input tokens is the accumulated context by the middle of an agent run: the system prompt, the tool definitions, and every tool result so far. Treat one call here as one turn of the loop, not one task.

Your numbers

Now put your own in

The fields open with the profile above. Change them to match your feature and the read updates as you type.

Your numbers

Rough numbers are fine. You are checking whether you are near the line, not being exact.

The usual right answer for production features. Near-frontier quality at a fraction of the cost.

Everything you send it

What it sends back

How often one user uses it

$
%

List prices as at 19 August 2026

What it costs you

Each use costs$0.108
Per user, per month$3.24
That is this much of what they pay you1.6%
You can afford to spend$59.70

Using 5% of what you can afford

The full answer

It pays for itself. See how much room you actually have.

  • The point at which this stops paying for itself
  • What it costs if people use it 2x, 5x or 10x more than you think
  • The same feature priced on every model, cheapest first
  • What to change, in the order worth changing it

We send you a copy, then roughly two emails a month. One click to stop.

What drives the cost

Accumulated context. An agent resends its entire history on every turn, so cost grows with the square of the run length rather than linearly. A ten-turn task costs far more than ten single calls, and this is the arithmetic that surprises people most.

Where the estimate goes wrong

Counting tasks instead of turns. Ask how many model calls one completed task actually takes, then multiply. A task that quietly takes fifteen turns rather than the five you assumed is three times the cost you budgeted, and nothing in the product surface tells you it happened.

What this does not price

Agent runs have a long tail: most finish quickly and some loop. Without a hard cap on turns, the tail sets your bill, not the median. Before building, decide what the agent does when it has not finished after N turns, and price the cap rather than the average.

Other features people cost

Next step

If multi-step agent workflow is on your roadmap and the number here made you pause, that is exactly the conversation worth having before an engineer starts.

Book the call

A 30-minute call. No deck.