AI / ECONOMICS

What does one successful AI task actually cost?

Change the inputs. Follow the arithmetic. All defaults below are illustrative, not live provider prices or benchmark results.

Tasks per monthAverage attempts per task (including retries)Input tokens per attemptOutput tokens per attemptInput price per 1 million tokens ($)Output price per 1 million tokens ($)Final task success rate (%)Other monthly costs ($)

Monthly model cost$96.00

Total modeled monthly cost$246.00

Cost per successful task$0.03

The model

Monthly model cost = tasks × attempts per task × ((input tokens × input rate) + (output tokens × output rate)) ÷ 1,000,000.

Cost per successful task = (monthly model cost + other monthly costs) ÷ (tasks × final success rate).

Assumptions and exclusions

Token use is averaged across attempts. Final success is measured at the task level, after retries. Put retrieval, tools, hosting, evaluation and human review into “other costs” where relevant. Caching, batch discounts, tiered pricing, latency penalties and several-model workflows are not separately modeled.

A zero final success rate makes cost per successful task undefined. Enter your own provider rates and observed success rate before using the result for planning.