BUYER GUIDE
AWS Glue Data Quality Pricing: AWS Glue Data Quality Cost per Table and per Run
AWS prices Glue Data Quality at 0.44 dollars per DPU-hour, and that single figure is where most cost estimates stop. The bill you actually get depends on four things the rate does not show: how many tables you check, how often, how many statistics each ruleset gathers, and whether anomaly detection is on. This page works the arithmetic from the examples AWS publishes on its own pricing page, so you can put a monthly number on your Redshift or S3 estate before you build a single ruleset, and compare it with a flat monitoring plan from 99 dollars a month.
14-day trial, no credit card, read-only connection
Alerted #data-eng 0.8s ago.
Downstream impact · consumers at risk
How much does AWS Glue Data Quality cost?
AWS Glue Data Quality costs 0.44 dollars per DPU-hour on the Glue pricing page, with a 2 DPU minimum and a 1 minute billing minimum for Data Catalog tasks. Anomaly detection adds 1 DPU per statistic for the 10 to 20 seconds each one takes. AWS prices its own example, 10 rules and 10 analyzers on one table, at about 0.081 dollars a run. Run that daily on 250 tables and the bill is roughly 608 dollars a month. Run it hourly and it is roughly 14,580. There is no license fee and no annual contract, but every table needs a ruleset someone writes and maintains. DataObservability monitors up to 250 Redshift tables for 299 dollars a month billed yearly, with no per-run charge.
Side by side
AWS Glue Data Quality pricing compared
Swipe to see all columns →
| Charge | Rate or figure | When it applies | What it means for the bill |
|---|---|---|---|
| Data quality task (Data Catalog) | 0.44 USD per DPU-hour | Every scheduled ruleset evaluation | Minimum 2 DPU, billed per minute with a 1 minute floor |
| Rule recommendation task | 0.44 USD per DPU-hour | Each time you ask Glue to draft rules for a table | AWS example: 5 DPU for 10 minutes = 0.37 USD per table |
| Anomaly detection | 1 DPU per statistic | Evaluation runs with anomaly detection on | 10 to 20 seconds per statistic, on top of the run itself |
| Retraining after an excluded anomaly | 1 DPU per statistic | When you exclude a data point to keep the model honest | About 15 seconds per statistic, AWS example 0.00185 USD |
| Checks inside a Glue ETL job | 0.44 USD per DPU-hour | Data Quality transform in a Spark job | Adds runtime to a job you already pay for |
| Checks inside a Flex ETL job | 0.29 USD per DPU-hour | Non SLA sensitive ETL jobs only | Flex is not supported for Data Catalog evaluations |
| Statistics storage | No charge | Every rule and analyzer run | Capped at 100,000 statistics per account, kept up to two years |
| Results and temporary files | Standard Amazon S3 rates | Every run that writes results | Billed under S3, not under Glue |
| Data Catalog | Free to 1 million objects and 1 million requests a month | Rulesets must attach to a cataloged table | Then 1 USD per 100,000 objects; crawlers bill DPU-hours |
| DataObservability Starter | 99 USD a month, billed yearly | 50 tables, one warehouse | No per-run or per-statistic charge, 7 day history |
| DataObservability Team | 299 USD a month, billed yearly | 250 tables, PagerDuty, ML anomaly detection | One flat figure regardless of check frequency, 90 day history |
| DataObservability Scale | 799 USD a month, billed yearly | 1,500 tables, column-level lineage, SSO | Multi-warehouse, audit log included |
Positioning and pricing models are summarized in good faith from each vendor's public pages, September 2026. Verify current terms with the vendor.
What you get
What AWS Glue Data Quality costs, line by line
The rate is simple and the bill is not
Glue Data Quality has one published rate, 0.44 dollars per DPU-hour, and AWS notes that pricing can vary by Region. The difficulty is that a DPU-hour is a unit of compute time, not a unit of data quality. What you pay depends on how long each evaluation run takes, how many DPUs you provision for it, how many statistics its rules and analyzers gather, and how many times a day you run it. None of that is visible when you sign up, because there is nothing to sign. You find out from Cost Explorer after the first month. The practical fix is to model the bill per table per run before you build rulesets, which is what the worked figures on this page do.
Start from the example AWS publishes, then multiply
The most useful number on the AWS Glue pricing page is buried in its fourth data quality example. A Data Catalog table with 10 rules and 10 analyzers gathers 20 statistics. The evaluation runs for 3 minutes on 2 DPUs, which costs 0.044 dollars. With anomaly detection on, AWS adds 1 DPU per statistic for about 15 seconds each, another 0.037 dollars, for a total of 0.081 dollars per run. That is a realistic per-table unit. Multiply it by tables and by runs per month. Daily checks on 50 tables cost about 122 dollars a month. Daily on 250 tables, about 608 dollars. Hourly on 100 tables, about 5,832 dollars. Hourly on 250 tables, about 14,580 dollars. Frequency is the variable that moves the bill most, and it is the one data teams most want to raise after the first incident.
The costs that do not show up under Glue
Three charges land on other lines of the AWS bill. Results, temporary files and shuffle files are written to an S3 bucket you choose and bill at standard S3 rates. Rulesets must be associated with a table in the AWS Glue Data Catalog, which is free up to 1 million objects and 1 million requests a month and then metered, and crawlers that keep the catalog current bill their own DPU-hours. And when the table lives in Redshift, the evaluation reads data out of the warehouse, so the queries it issues consume capacity on your Redshift cluster or Serverless workgroup. A fair total cost for Glue Data Quality adds all three to the DPU line.
The cost nobody puts in the estimate: writing rules
Glue Data Quality checks what you tell it to check. A ruleset is a DQDL document attached to one cataloged table, and it can hold up to 2,000 rules and 65 KB. Rule recommendations produce a draft, but somebody has to review it, tune thresholds, add the business rules the recommender cannot know, and keep the ruleset in step with schema changes. For a team with 20 critical tables that is a few days of work. For 250 tables it is a standing commitment. Put an hourly rate on that time and it often exceeds the DPU bill, which is the real comparison with a monitoring tool that learns each table baseline without a ruleset.
How it works
From connected to caught
Count the tables that actually need checking
List the Redshift or S3 tables that feed dashboards, finance reports or customer-facing exports. That count, not the size of the warehouse, drives both a Glue Data Quality bill and any monitoring plan.
Decide the check frequency per table
Most teams need daily checks on slow-moving tables and hourly checks on the few that load continuously. Price each group separately at about 0.081 dollars per run with anomaly detection on, using the AWS example as the unit.
Add the hours to write and maintain rulesets
Estimate the engineering time to review recommendations, add business rules and keep rulesets current as schemas change. Multiply by a loaded hourly rate and add it to the DPU figure.
Trial a flat plan on the same tables
Start a 14 day trial, connect read-only to Redshift, and let freshness, volume, schema and distribution monitors generate on every table. Compare what each approach caught and what each would cost over twelve months.
Worked monthly costs from the AWS example
The table below the fold on the AWS pricing page gives one realistic unit: 0.081 dollars for a 3 minute evaluation of one table with 20 statistics and anomaly detection on, or 0.044 dollars with anomaly detection off. Scaled across a month of 30 days, daily runs cost 2.43 dollars per table with anomaly detection and 1.32 without. Hourly runs cost 58.32 dollars per table with anomaly detection and 31.68 without. So 50 tables checked daily come to about 122 dollars a month, and 250 tables checked daily come to about 608. At hourly frequency the same 250 tables come to about 14,580 dollars a month with anomaly detection, or 7,920 without. Rule recommendation runs add a one-time charge on top, about 0.37 dollars per table in the AWS example, so roughly 93 dollars to draft rules for 250 tables. These figures are an extrapolation of AWS published examples, not a quote. Your runs may be shorter on small tables or longer on wide ones, and you may provision more than the 2 DPU minimum. Use them as a planning range, then check Cost Explorer after the first full month.
The minimum charge per run is small, and it is still a floor
A Data Catalog data quality task needs at least 2 DPUs and bills a minimum of 1 minute. At 0.44 dollars per DPU-hour that floor is about 0.015 dollars per run. It sounds negligible until you multiply it. A 250 table estate checked every hour runs 180,000 evaluations a month, so the floor alone is about 2,640 dollars before a single statistic is gathered or an anomaly is scored. Real evaluations rarely finish inside a minute, because each one starts compute, reads the table and writes results, which is why the AWS example uses 3 minutes.
Anomaly detection is billed per statistic, so wide rulesets cost more
When anomaly detection is on, AWS charges 1 DPU per statistic for the time it takes to score it, which it puts at 10 to 20 seconds on average. Every rule and analyzer gathers at least one statistic: a row count, a column count, a completeness percentage, a mean. A ruleset with 5 statistics adds very little. A ruleset with 60, which is common once analyzers cover every important column, adds about 0.07 to 0.15 dollars per run in anomaly scoring alone. Excluding an anomaly so the model does not learn from it triggers a retraining charge at the same 1 DPU per statistic for about 15 seconds. Anomaly detection on Data Catalog tables became available in July 2026, when AWS added evaluation runs with ObservationScope set to ALL, so many older cost estimates do not include it at all.
The 100,000 statistics limit is a planning constraint
AWS stores data quality statistics at no charge, but it documents a limit of 100,000 statistics per account, retained for up to two years. That limit interacts with scale in a way cost calculators ignore. Twenty statistics per run across 250 tables checked daily is 5,000 statistics a day and about 150,000 a month. Hourly checks multiply that by 24. Before committing to Glue Data Quality across a large estate, ask your AWS account team how the limit applies to your account and whether it can be raised, and plan analyzer coverage around the answer.
ADVANCED recommendations and what the pricing page does not say yet
On September 18, 2026 AWS added an ADVANCED rule recommendation mode that uses Amazon Bedrock, table metadata and sampled rows to draft DQDL rules. BASIC stays the default. When we read the Glue pricing page in September 2026 it listed no separate rate for ADVANCED recommendations. If you plan to use it, confirm with AWS whether model usage bills on top of the recommendation task DPU-hours, and test it on a handful of tables before running it across the catalog.
Data Catalog checks versus checks inside ETL jobs
Glue Data Quality has two entry points and they price differently. Checks on Data Catalog tables run as separate tasks you provision and schedule, with the 2 DPU and 1 minute minimums. Checks inside a Glue ETL job run as part of the job, so they add runtime and DPU consumption to a job you already pay for, and they can use Flex at 0.29 dollars per DPU-hour for work that is not time sensitive. The ETL path is also the only one that identifies which records failed; for Data Catalog evaluations that is not supported. For Redshift tables loaded by something other than Glue, such as Fivetran, dbt or a COPY from S3, the Data Catalog path is usually the only one available.
When AWS Glue Data Quality is the cheaper answer, stated plainly
If you check fewer than about 40 tables once a day, Glue Data Quality will usually cost less per month than any paid monitoring plan, including ours. It also wins when the checks are precise business rules you already know how to state, when you want to quarantine failing rows inside a Glue ETL job before they land, and when procurement makes any new vendor slow. It stays inside your AWS bill, your IAM roles and your CloudFormation. Those are real advantages and a monitoring tool does not replace them.
Where a flat monitoring plan comes out ahead
The balance flips as tables, frequency and statistics grow, and it flips hardest on coverage. Glue Data Quality only watches the tables someone wrote a ruleset for. DataObservability connects to Redshift read-only, learns a baseline for freshness, volume, schema and distribution on every table from its own history, and alerts in Slack on every plan and PagerDuty from Team up. Starter is 99 dollars a month billed yearly for 50 tables, Team is 299 for 250 tables with end-to-end lineage and ML anomaly detection, and Scale is 799 for 1,500 tables with column-level lineage, SSO and an audit log. The price does not change when you check more often, and nobody writes a ruleset per table. Many teams run both: Glue Data Quality for the handful of business rules only they can state, and a monitor for everything else.
Questions buyers ask
AWS Glue Data Quality pricing FAQ
Is AWS Glue Data Quality free?
No. There is no license fee or annual contract, but every recommendation task, evaluation run and anomaly detection step bills at 0.44 dollars per DPU-hour. Storing statistics is free up to the 100,000 per account limit. Results written to S3 bill at S3 rates, and Data Catalog usage beyond its free tier is metered.
What is the minimum charge for an AWS Glue Data Quality run?
Data Catalog data quality tasks require at least 2 DPUs and bill a minimum of 1 minute, so the floor is about 0.015 dollars per run at 0.44 dollars per DPU-hour. Most runs take longer. AWS prices a 3 minute run on one table at 0.044 dollars without anomaly detection.
Does AWS Glue Data Quality anomaly detection cost extra?
Yes. With anomaly detection on, AWS adds 1 DPU per statistic for the 10 to 20 seconds it takes to score each one. In the AWS example, 20 statistics add about 0.037 dollars to a 0.044 dollar run, nearly doubling it. Retraining after you exclude an anomaly bills the same way.
How much does AWS Glue Data Quality cost for 100 tables?
Using the AWS example of 0.081 dollars per run with 20 statistics and anomaly detection on, 100 tables checked once a day cost about 243 dollars a month. Checked hourly, about 5,832 dollars a month. Add S3 storage for results and the engineering time to write and maintain 100 rulesets.
Can I use Glue Flex to cut data quality costs?
Only for checks inside Glue ETL jobs. Flex runs at 0.29 dollars per DPU-hour instead of 0.44 for work that is not time sensitive. Flex is not supported for data quality evaluations on Data Catalog tables, which is the path most Redshift tables use.
Is AWS Glue Data Quality cheaper than a data observability tool?
For a small number of tables checked daily, usually yes. At about 40 or more tables checked daily, or any sizable estate checked hourly, the DPU bill passes a flat plan such as 99 dollars a month for 50 tables or 299 for 250, before counting the time spent writing rulesets.
Does the Glue pricing page list a rate for ADVANCED rule recommendations?
Not when we read it in September 2026. AWS added ADVANCED recommendations, which use Amazon Bedrock, on September 18, 2026, and the pricing page still described recommendation tasks only in DPU-hours. Confirm with AWS whether model usage bills separately before running it across your catalog.
Does AWS Glue Data Quality work with Redshift?
Yes, through the Data Catalog. The Redshift table must be cataloged, a ruleset attached to it and an evaluation run scheduled. The evaluation reads from the warehouse, so it uses Redshift capacity as well as Glue DPU-hours, and identifying individual failed records is not supported on this path.
More of the platform
Catch broken data before your stakeholders do
Connect your warehouse and get aws glue data quality pricing live from one read-only connection. Transparent pricing, no credit card.