Datafold vs dbt tests

Quick verdict

dbt data tests are the built-in testing feature of dbt. A data test is a query that returns failing records: zero rows means it passes. dbt ships four generic tests (unique, not_null, accepted_values, relationships), you can write your own generic tests or install packages, and singular tests are one-off SQL files in the tests directory. Severity can be warn or error, with warn_if and error_if thresholds, and store_failures saves failing rows to an audit table. Unit tests, added in v1.8, check model logic on static inputs before the model is built.

Datafold is a commercial data engineering platform built around data diffing: comparing tables, query results or files row by row and column by column, within one database or across two. It runs scheduled diff and data test monitors, offers ML-driven anomaly detection monitors, and in CI diffs the dbt models a pull request changes. The company now leads with its Migration Agent, which combines AI code translation with diff-based validation, and exposes diffs to coding agents through an MCP server.

Side by side

Key differences

Assertions versus diffs

The two are complementary rather than competing. A dbt data test checks a rule you already know. For example, a singular test that fails if any order has a negative total:

-- tests/assert_no_negative_order_totals.sql (dbt singular data test)
select order_id, sum(amount) as total_amount
from {{ ref('fct_payments') }}
group by 1
having sum(amount) < 0

If a refactor silently drops 2 percent of rows, or changes a currency conversion so every value shifts slightly, no test fails unless someone anticipated that exact failure. Datafold catches that class of problem by comparing the production version of each changed model with the pull request version and reporting added, removed and changed rows and values. The trade-off is cost and setup: diffs need two builds and a paid service, while tests need only dbt.

How each works in CI

With dbt, CI usually builds the changed models and their children in a temporary schema and runs their tests, with unit tests recommended for development and CI rather than production. Any CI system that can run the dbt CLI works, and the same commands run on the dbt platform. With Datafold, two manifest.json files (production and pull request) are submitted; dbt Cloud does this automatically, while dbt Core users add a step to their CI configuration (the documentation gives CircleCI, GitHub Actions and GitLab as examples). Datafold compares the manifests, diffs the modified models in the warehouse, identifies downstream assets such as BI tools, and posts a summary comment on the pull request, with an optional AI overview on GitHub, GitLab and Bitbucket.

The important caveat is availability: Datafold's documentation says CI/CD testing is available only in Datafold Classic and is not available in the current Datafold app (us.datafold.com, eub.datafold.com and dedicated or customer-hosted deployments). Ask Datafold which app a new contract would use and what the plan is for CI.

Migrations and cross-database checks

dbt tests run in one warehouse, against the project's own models. They cannot tell you whether a table in Teradata or Oracle matches its copy in Snowflake. Datafold diffs across databases and lists 23 connectors, including Teradata, Netezza, Db2, SAP HANA, Oracle, SQL Server, Snowflake, Databricks and BigQuery, which is why it sells migrations as an outcome with price, timeline and quality guaranteed in the contract. If a migration is your trigger for looking at Datafold, compare it with a dedicated source-to-target testing tool as well; see QuerySurge vs Datafold.

Cost and lock-in

dbt data tests cost nothing beyond the warehouse queries they run, and because tests are SQL in your repository they move with the project. Large suites do add query cost on every run, so many teams tag tests and run the expensive ones less often. Datafold is quote only, with no free tier, and the free data-diff tool that many teams used for ad hoc diffs is archived (the repository remains readable under MIT but is not maintained). Budget for a contract if diffing becomes part of your review process.

Pricing and licensing

Datafold. No published prices. datafold.com/pricing redirects to a contact page, and the home page offers a demo with a "discuss pricing" step. For data platform migrations Datafold says pricing is based on the number of objects, with price, timeline and quality guaranteed in the contract. No free tier or self-service trial is advertised.

dbt data tests. Included in dbt. With open source dbt Core (Apache 2.0) they are free; you pay only for warehouse compute. On the managed dbt platform, the pricing page lists Developer free (1 developer seat, 1 project, 3,000 successful models built a month), Starter at USD 100 per user per month (up to 5 developer seats, 15,000 models built a month, overage billed) and Enterprise and Enterprise+ by quote. Tests run on every plan.

Pricing checked on the vendors' official pages on 7 November 2026. Prices change; confirm before buying.

Where each one leads

Limitations

When to choose each

When neither is right

• You need ongoing production monitoring of freshness and volume: see Elementary vs Metaplane.
• Your pipelines are not dbt: see Soda vs Great Expectations.
• Your testers write paired source and target SQL for ETL and BI validation: see QuerySurge vs Datafold.
• You only need to compare two SQL Server databases once: a data compare utility such as Redgate SQL Data Compare may be enough.

Final recommendation

Frequently asked questions

Sources

• Datafold docs: how Datafold in CI works
• Datafold docs: database connections
• Datafold home page
• Archived datafold/data-diff repository
• Datafold blog: sunsetting open source data-diff
• dbt docs: add data tests to your DAG
• dbt docs: unit tests
• dbt pricing
• dbt-core repository and licence
• Fivetran press release: Fivetran and dbt Labs complete merger

Checked November 2026.

How we research comparisons: our editorial method.