Cost per successful task calculator
Agents retry, loop, and abandon. Cost per token misses that — this prices the task that actually finished.
Cost per token prices what the fleet produced. It says nothing about whether the work succeeded. Once agents are in the path — many dependent calls per task, retries on failure, loops that get abandoned — the real unit of cost is the completed task, and it can be several times the naive number.
01: The task, as designed
02: The task, as it actually runs
Where each task dollar goes
Green cells are spend that produced a finished task. Hollow cells are retry, loop, and abandonment waste.
More Info
Cost per token is a production metric; cost per successful task is a business metric. The gap between them is retry amplification—failed and repeated calls billing identically to useful ones—plus loop waste and abandoned chains whose entire spend is carried by the tasks that finish. As agents take over more of the traffic, this gap is where inference budgets quietly go.
Frequently asked questions
From AI factory math to AI factory control
Health-aware routing prevents the retries, per-tenant budgets stop the loops, and token attribution makes abandonment visible per agent—which is why the waste multiple is a traffic control number, not a model quality number.