AWS Budgets compares costs, usage and the utilisation and coverage of reservations and Savings Plans with thresholds and alerts when the actual value or the forecast exceeds them; monitoring costs nothing, budget actions cost per day from the third action-enabled budget, and budget reports per delivery. Cost Anomaly Detection learns from the net costs per service, account, tag or category and reports outliers with their cause by service, account, region and usage type, without a fee of its own. Both read the same data as Cost Explorer, with up to 24 hours of delay and up to three updates a day, and that is the limit both share: a budget reports an overrun hours after the usage, not before.
Pricing dimensions
The two services cost almost nothing, and what they cost sits in two rows; the other rows of the table are the deadlines and limits that decide when an alert comes. Every cost line a budget reads carries its usage type, and budgets filter by it.
| Dimension | Unit | What drives it | Optional |
|---|---|---|---|
| Budget monitoring | Free; cost, usage, RI and Savings Plans budgets with thresholds on actual and forecast, per month, quarter or year | Every budget, its filters and notifications by email and SNS | Yes |
| Budget actions | The first two action-enabled budgets a month free, whatever the number of their actions; every further one USD 0.10 per day | Budgets with an IAM policy, an SCP or stopping named EC2 and RDS instances as the action | Yes |
| Budget reports | USD 0.01 per delivered report, daily, weekly or monthly by email | Every report on a budget portfolio | Yes |
| Refresh | Up to three times a day, typically 8 to 12 hours after the previous one; the delay between usage and billing comes on top | When a threshold counts as exceeded | No |
| Cost Anomaly Detection | Without a fee of its own; monitors by AWS services, member account, cost allocation tag or cost category; thresholds absolute or in percent | Every monitor and its alert subscriptions | Yes |
| Detection | About three times a day on net unblended costs; data with up to 24 hours of delay; a new monitor needs 24 hours, a new service ten days of history | When an anomaly is detected | No |
| Cost Explorer API | USD 0.01 per request with the primary billing view | Reading budgets, monitors and anomalies through the API | Yes |
| Purchase option | None |
A budget is a threshold on a number AWS recomputes up to three times a day: costs, at choice blended, unblended, net, amortized or net amortized, with or without discounts, credits, support and taxes; usage in a service's unit; or the utilisation and coverage of reservations and Savings Plans, where the alert comes when the value falls below the threshold. The sum applies per month, quarter or year, fixed or planned per period, filtered by account, service, region, usage type, tag or cost category, and the notification goes by email or to an SNS topic, on actual values or on the forecast. An action turns the alert into an intervention: an IAM policy or a service control policy that prevents new resources, or stopping named EC2 and RDS instances in the own account, automatically or after approval; the management account can apply an SCP to a member account but not stop instances there. The mechanics of thresholds across all providers are in budgets and budget alerts and thresholds; what a reservation's utilisation means, in commitment coverage and utilisation.
Cost Anomaly Detection needs no threshold on a sum but learns the pattern: a monitor by AWS services, managed by AWS, tracks the up to 5,000 largest values of a dimension one by one; a monitor by member account, cost allocation tag or cost category, managed by the customer, tracks the chosen values as a sum. Detection runs about three times a day on net unblended costs after all discounts, with Cost Explorer's data and its delay of up to 24 hours, and a detected anomaly names its causes by service, account, region and usage type, ranked by amount. An alert subscription sets the threshold, absolute in USD or in percent of the expected value, and the channel: individual alerts through SNS, daily summaries of the ten largest anomalies of the previous day by email at 00:00 UTC, or weekly ones. Marketplace products other than third-party models on Bedrock are not monitored, and an account whose bill another account pays through billing transfer cannot use the service. How anomalies are detected in a tool outside AWS is described in anomaly detection.
Invoices, exports and APIs
Budgets and monitors do not sit in the Cost and Usage Report; they read it. What sits on the invoice are the fees: the action-enabled budgets from the third per day and the budget reports per delivery as lines of the service AWS Budgets in the management account, with the product code AWSBudgets, and the Cost Explorer API requests under AWSCostExplorer. A budget reads the same data as Cost Explorer in the chosen cost type, and the delay between usage, billing and budget is the one data freshness in reports describes: an instance that starts at 09:00 appears hours later in the report and in the budget's next run. A budget's forecast comes from Cost Explorer's forecast, which needs history; an account without a past gets no forecast alerts.
The Budgets API, CreateBudget, DescribeBudgets and DescribeBudgetPerformanceHistory, manages budgets and delivers their history per period; the Cost Explorer API delivers the anomalies and monitors with GetAnomalies, GetAnomalyMonitors and GetAnomalySubscriptions, at USD 0.01 per request. An alert through SNS reaches EventBridge, chat applications through Amazon Q Developer and Lambda functions of your own. Member accounts create budgets of their own and see only those; a management account's budget on a member account is invisible to the member, because budgets are not shared across accounts (Organizations and consolidated billing). The budgets for the free tier of the first twelve months are switched on for the organisation by the management account (free tiers and trials). Which filters a budget gets depends on the tags and categories that are activated (cost allocation tags and Cost Categories).
Levers
Configuration: the first lever is the number of action-enabled budgets, because only they cost: two a month are free, and a budget with several actions counts as one, so the actions belong on few budgets instead of many. The second is the threshold: a budget on the forecast alerts before the month's end, one on the actual value after, and both together with steps at 80, 100 and 120 percent replace many budgets on one sum. The third is the monitors: the AWS-managed monitor by services covers an account without upkeep; a monitor per team by cost category reaches the people responsible, and the threshold in percent instead of absolute scales with the team.
Purchase option: none; for reservations and Savings Plans the utilisation and coverage budgets are the control, not a purchase.
Architecture: budgets alert, actions intervene, and both come hours after the usage; a limit that is to hold before the usage is a service quota, an SCP or a guardrail in the deployment, as guardrails and approval thresholds describes. The budgets in Costfluent run on the report and are independent of the budgets in AWS; what carries a forecast and how a variance is explained is in forecasting methods and variance analysis.
Where it goes wrong
- The budget alerts after the money is gone. Up to three times a day, 8 to 12 hours after the previous refresh, plus the billing delay. A budget on the forecast and an action at 90 percent alert earlier; a service quota holds before.
- The action does not stop the wrong instance. An action stops only named instances in the own account; in member accounts the management account can only apply an SCP. The SCP prevents new things, it ends nothing.
- Every team has its action-enabled budget. Twenty action-enabled budgets cost eighteen times USD 0.10 per day. Actions go on the account's budget, not on every service budget.
- The anomaly comes a day late. Detection reads data with up to 24 hours of delay and runs three times a day. A Monday morning anomaly sits in Tuesday's daily summary.
- The new service is not detected. A service needs ten days of history, a monitor 24 hours. A usage budget watches a new service's first week.
- The absolute threshold fits only one team. USD 100 is noise for one team and the month for another. The threshold in percent of the expected value per cost category fits both.
- The Marketplace invoice goes unnoticed. Cost Anomaly Detection does not monitor Marketplace products. A cost budget with the filter on the billing entity AWS Marketplace reports them.
Example
List prices in USD, read on 2026-10-21: budget monitoring and Cost Anomaly Detection cost nothing, an action-enabled budget from the third USD 0.10 per day, a budget report USD 0.01 per delivery, a Cost Explorer API request USD 0.01.
An organisation with twelve accounts runs a cost budget per account with alerts at 80 percent of the forecast and 100 percent of the actual value, plus five action-enabled budgets that apply an SCP against new instances at 100 percent: monitoring costs nothing, the three action-enabled budgets above the two free ones 3 × 30 × 0.10 = USD 9.00 a month, a daily report on the portfolio 30 × 0.01 = USD 0.30. A dashboard that reads the budgets and anomalies every hour through the API, with twelve requests per run, pays 12 × 24 × 30 × 0.01 = USD 86.40, nine times more than the budgets themselves. A monitor by cost category per team with a threshold of 10 percent reports an outlier from about USD 67 a day on a team with USD 20,000 a month, from about USD 7 on one with USD 2,000, where an absolute threshold of USD 100 would never have reached the small team. An instance that starts on Monday at 09:00 and costs USD 50 a day sits at the earliest in Monday evening's budget run and in Tuesday's daily summary; by then USD 50 to 100 are spent, and a budget on the forecast would have reported the month's overrun the same evening. The figures show that both services' fees vanish beside the API, that the threshold in percent makes the monitors per team comparable, and that a budget alerts hours, not minutes, after the usage.
Sources
Read on 2026-10-21. Prices: AWS Budgets pricing, AWS Cost Explorer pricing. Budgets: managing your costs with AWS Budgets, configuring budget actions. Anomalies: detecting unusual spend with AWS Cost Anomaly Detection, getting started with AWS Cost Anomaly Detection, Cost Explorer users can now automatically detect cost anomalies at no additional cost. Organisation: free tier for AWS Organizations and volume discounts.