dbt State is now generally available everywhere dbt runs, including Airflow, Dagster, GitHub Actions and laptops, and on Snowflake's dbt projects. It supports Snowflake, BigQuery, Databricks and Redshift.

Instead of running dbt jobs on a schedule, dbt State runs them on change. Each run reads model SQL and warehouse metadata, decides whether a model's result would differ, then builds, skips, clones or defers each node. No selection syntax, manifest scripts or custom orchestration are needed.

The idea is simple: pipelines do a lot of redundant work, and reusing what hasn't changed saves run time and warehouse compute. dbt Labs says more than 22 million models are built on dbt daily, and early adopters have averaged 15-30% compute savings.

RxBenefits Principal Data Engineer Chris Shepherd said scheduled jobs on a Snowflake adaptive warehouse cut costs by 59%, with over 700,000 models reused instead of rebuilt, reducing query run time by two weeks over a 60-day period.

Virgin Media O2's Gordon Curzon described a paradigm shift, citing 25% savings on job run time and BigQuery compute costs. Custom Ink's Brett Petersen said a previous setup was simplified to one run.

Freshness now lives in the model rather than the job schedule, via a lag_tolerance setting, so dbt State decides per model, per run, whether requirements are met.

In development, dbt State handles cloning and reuse automatically. Joe & The Juice Lead Data Engineer Mykkel Rydal said iteration on a 400M+ row sales fact table that took 15-25 minutes now resolves in seconds.

Pricing is consumption-based, tied to daily active target tables, with every reuse after the first in a day free. A 30-day free trial is available.

Source: dbt Labs Blog