Data Profiling Tools: The Best Data Profiling Software Compared for 2026

One-off exploratory profiling of a pandas or Spark dataframe

This is the package the whole category still calls pandas-profiling, and its name has now changed twice. It became ydata-profiling in 2023, and in April 2026 it was renamed again to fg-data-profiling, with the repository moving to the Data-Centric-AI-Community organization. The PyPI listing for the old ydata-profiling package now carries an explicit notice that it will no longer receive updates or bug fixes. On capability it remains the best free exploratory profiler there is: one line of code produces an HTML report with inferred types, missing-value maps, distributions, correlations, interactions, and automatic warnings, for pandas and Spark dataframes. Two things to weigh. It loads data into your process, so it profiles a sample rather than a billion-row warehouse table, and the renamed package has published only two releases, both on 22 April 2026, with no repository commits since. Check its current state before you standardize on it.

Teams that want profiling and checks in one declarative YAML file

Soda splits the difference between profiling and testing. SodaCL is a YAML checks language that includes profiling directives, so you can ask for column profiles and distribution statistics in the same file that holds your pass or fail assertions, and it pushes the work down into the warehouse rather than pulling rows out. Soda Core is open source and actively released, with 4.19.0 published on 28 July 2026. Soda is also one of the few vendors in this whole space that publishes prices: a free tier and a Team plan listed at $750 per month, with enterprise quoted. The trade is that profiling is a feature of a testing product rather than the point of it, so the reports are less exploratory than a dedicated profiler and more oriented toward monitoring the same columns repeatedly.

dbt projects that want profile stats as a model or in docs

A small dbt package that generates profile statistics for a relation as SQL, so the profile is computed by the warehouse and can be materialized as a model or dropped straight into your dbt docs. For a team already living inside dbt this is the lowest-friction option on the page: no new service, no data leaving the warehouse, and the output sits in version control with everything else. Keep expectations proportional to its size. It is a community package with a small maintainer base, last updated in February 2026, it profiles a relation you point it at rather than the whole estate, and there is no UI, no history, and no alerting. It answers what does this table look like right now, not what changed since last week.

Writing and running data quality assertions, not profiling

Great Expectations appears on nearly every data profiling list, and that placement is now out of date. The profiling helpers those lists point at, the UserConfigurableProfiler and the onboarding data assistants, belonged to the 0.18 line. GX 1.0 shipped in August 2024 as a hard breaking change, and one of its own changelog entries records removing all mentions of profilers from the public docs. What GX is today is an excellent assertion framework: you declare Expectations, group them into suites, and run them through Checkpoints. That is validation, which starts from a rule you already know. Profiling is the step before it, which is how you find out what the rules should be. Ownership also split in 2026, with FICO acquiring GX Cloud and Fivetran becoming steward of GX Core, so evaluate the open-source library rather than the hosted product.

AWS estates wanting no-code profile jobs on S3 and Redshift

DataBrew is the no-code option inside AWS. A profile job points at a dataset in S3, Redshift, or a JDBC source and produces a full statistical summary without any code: inferred types, missing and duplicate counts, value distributions, correlations, and standard summary statistics, with results written back to S3 and viewable in the console. Its real advantage is that it publishes prices, which almost nothing else in this category does. Interactive sessions are $1.00 per thirty minutes and jobs run at $0.48 per node hour, billed by the minute, so you can model the cost of profiling your estate before talking to anyone. It is AWS-only, and it is a batch job rather than continuous monitoring, so pair it with something that watches for drift if that is the actual need.

BigQuery estates wanting scheduled profile scans

Google runs data profiling as a scheduled scan against a BigQuery table, publishing per-column statistics and feeding the results into data quality rule recommendations. Note the name: this was Dataplex Universal Catalog until 10 April 2026, when it was renamed Knowledge Catalog, with the API, CLI, and IAM names left unchanged and the billing SKUs still saying Dataplex. Pricing is published and specific, which is welcome, but read the tier carefully because it catches people out. Profiling sits in the premium processing tier at $0.089 per DCU-hour, and the 100 free DCU-hours per month apply only to the standard tier at $0.06. Profiling has no free allowance at all, and consumption scales with rows, columns, bytes scanned, and how often you schedule the scan.

Databricks teams wanting profiles tracked over time

The most complete native profiler of the four platforms, largely because it stores history instead of printing a report. A monitor writes a profile metrics table you can query like any other, holding count, num_nulls, avg, min, max, stddev, distinct_count, median, a 1,000-value quantile array, num_zeros, num_nan, the percent variants of those, string length statistics, and the top 100 frequent items per column. Three profile types cover the common shapes: Snapshot for a table as it stands, TimeSeries for metrics sliced by a timestamp column and granularity, and InferenceLog for model monitoring, which adds accuracy and fairness metrics. Because the output is a table, drift is a query rather than a feature request. It requires Unity Catalog and stops at the Databricks boundary.

Snowflake shops wanting scheduled column metrics natively

Snowflake exposes profiling as data metric functions, which are metrics attached to a table or column and evaluated on a schedule, with results landing in an event table you can query and alert on. System functions cover null counts, duplicates, uniqueness, freshness, and row counts, and you can write your own in SQL. Four details decide whether this is enough. It requires Enterprise Edition or above. Creating a DMF or calling one inside a SELECT is not billed, and only a scheduled DMF meters serverless compute. The default schedule is hourly. There is a hard cap of 50,000 DMF associations per account, and DMFs cannot be set on a hybrid table or a stream. Inside those limits it is the cheapest credible starting point for a Snowflake-only team.

Large estates already standardized on Informatica

The deepest profiling engine in the category, and the oldest. Informatica Data Quality does column profiling, cross-column dependency and key discovery, and cross-source relationship discovery, which is the analysis that finds an undeclared foreign key between two systems nobody documented. If your problem is a migration where the source schemas are a mystery, this class of tool exists for exactly that. Salesforce completed its acquisition on 18 November 2025, and the product now ships as Informatica from Salesforce inside IDMC. Pricing runs on consumption units called IPUs; the model is documented but Informatica publishes no dollar rate per IPU. Hybrid and on-premises sources need a Secure Agent sized at 16GB of RAM minimum, so factor the infrastructure in.

Mainframe and legacy estates in a formal data program

The reference implementation of classical data profiling, and still the answer for the estates it was built for. Information Analyzer performs column analysis, primary key analysis, natural key analysis, and cross-domain analysis to find overlapping and redundant data across sources, with the results feeding a formal data quality program. It reaches sources the cloud-native tools do not, including mainframe and long-lived enterprise databases, which is usually the deciding factor. It is heavyweight to deploy and operate, the interface shows its age, and there is no published pricing. Choosing it in 2026 for a Snowflake and dbt stack would be an odd decision; choosing it for a thirty-year-old estate under audit is entirely reasonable.

Quality-led governance programs wanting profiling built in

Ataccama bundles profiling, data quality, master data management, and governance into one platform, and profiling is the front door: it scans sources, builds profiles, and proposes quality rules from what it finds rather than making you write them from a blank page. That autogeneration is the genuine differentiator when nobody on the team can yet say what good looks like. It runs as SaaS or self-hosted, and it markets a Leader placement in the 2026 Gartner Magic Quadrant for Augmented Data Quality Solutions. It publishes no pricing, and the third-party annual figures circulating for it do not come from Ataccama, so do not plan a budget around them. It is a platform purchase, not a tool you adopt on a Tuesday.

Talend estates, but see the Open Studio warning below

Talend built a well-regarded profiling and cleansing suite, and it still exists inside Qlik, which acquired Talend in May 2023. The warning is about the free edition that open-source profiling lists still recommend. Talend Open Studio reached end of life at version 7.3 on 31 January 2024, and the downloads came down that day. We followed the old product URL on 5 August 2026 and it now redirects to a commercial Qlik Talend Cloud page. There is no free Talend edition to download, whatever the roundups say. The paid product is capable, profiling is genuinely good, and Qlik markets a Leader placement in the 2026 Gartner Magic Quadrant for Augmented Data Quality Solutions. It publishes no pricing.

Microsoft estates wanting profiling inside governance

Purview folds profiling into its data quality capability inside Unified Catalog, so profiles run against assets that are already governed rather than as a standalone exercise. If your estate is Microsoft, that shared control plane across catalog, quality, and compliance is the argument, and Microsoft is one of the very few vendors here that publishes exact metered rates. Governance is $0.0165 per governed asset per day plus $15, $60, or $240 per data governance processing unit, and Microsoft documents roughly 0.02 processing units per simple rule run against a million rows. A billing detail that saves real money: an asset only counts once it is linked to a governance concept, so assets merely scanned into the Data Map are not billed. Keep the three Purviews apart, since the classic Data Catalog is closed to new customers and Azure Data Catalog was retired in May 2024.

Databricks Lakehouse Monitoring

The most complete of the four, because it keeps history. A monitor writes a profile metrics table holding count, null count, average, min, max, standard deviation, distinct count, median, a 1,000-value quantile array, zero and NaN counts, the percent variants, string length statistics, and the top 100 frequent items per column. Three profile types cover the usual shapes: Snapshot, TimeSeries sliced by a timestamp column, and InferenceLog for model monitoring. Because the output is an ordinary Unity Catalog table, detecting drift is a SQL query rather than a feature request. It requires Unity Catalog, and it stops at the Databricks boundary.

Snowflake data metric functions

Snowflake models profiling as metrics attached to a table or column and evaluated on a schedule, with results written to an event table you can query and alert on. System DMFs cover null counts, duplicates, uniqueness, freshness, and row counts, and custom ones are ordinary SQL. Enterprise Edition or above is required. Creating a DMF or calling it inside a SELECT is not billed, and only a scheduled DMF meters serverless compute, which makes ad hoc use effectively free. The default schedule is hourly, there is a cap of 50,000 associations per account, and DMFs cannot be attached to a hybrid table or a stream.

Google Knowledge Catalog

Profiling runs as a scheduled scan against a BigQuery table, producing per-column statistics that also feed automatic data quality rule recommendations. The service was renamed from Dataplex Universal Catalog on 10 April 2026, with API, CLI, and IAM names unchanged and billing SKUs still carrying the Dataplex name. Pricing is published, which is rare, but check the tier: profiling is premium processing at $0.089 per DCU-hour and the 100 free DCU-hours per month apply only to the $0.06 standard tier. Profiling has no free allowance, and consumption scales with rows, columns, bytes scanned, and scan frequency.

AWS Glue DataBrew

The no-code option in AWS. A profile job points at S3, Redshift, or a JDBC source and returns inferred types, missing and duplicate counts, distributions, correlations, and summary statistics, written back to S3 and rendered in the console. It publishes exact prices, at $0.48 per node hour for jobs and $1.00 per thirty-minute interactive session, billed by the minute, so you can model the cost of profiling your estate up front. It is a batch job rather than a continuous monitor, so nothing watches for drift between runs unless you schedule and diff them yourself.