Guide

How do you calculate what an AI agent costs per month?

To calculate AI agent cost per month, multiply the number of runs by the cost of one run, then add people time and fixed costs. Measure the cost of a real run first, because an agent chooses how many steps to take. Use the worksheet below, and compare options on the same tasks.

  • Sourced
  • Updated 16 Nov 2026

Get Your Free Trial

In 30 seconds

Three numbers you need

Runs

How many a month.

Cost per run

Measured, not guessed.

Review time

People minutes per run.

The formula

Written out:

Monthly cost = (runs × cost per run) + (runs × review minutes × cost of a minute) + fixed costs

An AI agent cost calculator is only as good as its inputs. A price list does not give you the cost of a run. Measure it.

The worksheet

Fill in one row for each task your agents do.

TaskRuns a monthCost of one runReview minutes per runMonthly cost
Task 1ABCA × B + A × C × minute cost
Task 2
Task 3
Fixed costs (setup, tools)
Total

How to get the cost of one run

  1. Run the task five times on real input.
  2. Write down the cost shown for each run.
  3. Use the highest of the five, not the average, for your first budget. Agents vary.
  4. Re-measure after any change to the task or the model.

A worked example

This is an example with round numbers, not a price. Suppose a task runs 200 times a month, one run costs 4 units, and a person spends 2 minutes reviewing each run at 1 unit a minute. The runs cost 800 units and the review costs 400 units, so the task costs 1,200 units a month before fixed costs. If a limit stops the task at 1,000 units, the last runs wait for a person to decide.

Cost per task, and comparing options

LLM cost per task is the number to compare. On one public maths benchmark in November 2026 the top-scoring model cost about 15 times more per run than a cheaper one that scored a few points lower (The Decoder). Compare any two options on your own five tasks, not on a headline price. An AI agent cost comparison on the same tasks is fair. One on different tasks is not.

Set the limit before the month starts

A calculator gives you a number. A limit makes it true. Decide what happens at the limit: warn, ask a person or stop. See AI spend governance.

How MFDIO applies this today

MFDIO measures AI work in Work Units, one meter a business can plan with, and shows the cost of each request or run, so you can fill the worksheet from real runs. Limits warn, ask or stop, overage is off by default and an Emergency Stop pauses all AI work. See AI cost by team and pricing, including the free 14-day trial.

Questions people ask

Is there an interactive calculator? Not yet. This worksheet works with any tool.

What is a fair cost per run? There is none. It depends on the task. Measure it.

Who should own the number? Finance and operations together, with the owner of each agent.

Limits of this guide

This guide is general information and not legal advice. MFDIO does not hold SOC 2 or ISO 27001 today. Market figures are analyst or survey estimates and vary between sources.

Sources

  1. Gartner, June 2025: over 40% of agentic AI projects predicted to be canceled by the end of 2027, citing cost, unclear value and weak risk controls
  2. The Decoder, "Gemini 4 Carbon takes the MathArena crown but costs 15 times more than GPT-6.1 Sol", November 2026 (one benchmark; results vary by task)
  3. MFDIO product facts: checked against /how-it-works, /safe-by-design and /status on 16 November 2026

Published . Last updated . Written by the MFDIO team.

See Cost by Run

Free for 14 days, no card.

Get Your Free Trial