Databricks Release Notes

New columns are now available in the query history system table, providing additional query insights:

• cache_origin_statement_id: For query results fetched from cache, this field contains the statement ID of the query that originally inserted the result into the cache.
• query_parameters: A struct containing named and positional parameters used in parameterized queries.
• written_rows: The number of rows of persistent data written to cloud object storage.
• written_files: Number of files of persistent data written to cloud object storage.

See Query history system table reference.

Tables created by Spark Declarative Pipelines on Lakeflow in Unity Catalog ETL pipelines can be moved from one pipeline to another. See Move tables between pipelines.

Inline execution history is now available in the new SQL editor, allowing you to quickly access past results without re-executing queries. Easily reference previous executions, navigate directly to past query profiles, or compare run times and statuses—all within the context of your current query.

INSERT REPLACE USING is now in Public Preview for Databricks Runtime 16.3. This SQL command replaces part of the table with the result of a query. Rows are replaced when all columns listed in USING match with the = operator.

For Public Preview, columns in the USING clause must be the full set of the table's partition columns.

See INSERT in the SQL language reference and Selectively overwrite data with Delta Lake.

INSERT REPLACE ON is now in Public Preview for Databricks Runtime 17.1. This SQL command replaces part of the table with the result of a query. Rows are replaced when they match the user-defined condition.

See INSERT in the SQL language reference.

In Beta, Supervisor Agent supports designing a multi-agent AI system that combines Genie Agents (formerly Genie Spaces) and Knowledge Assistant agent endpoints to work together on complex tasks that require different specialized skills. Supervisor Agent integrates with the Review App, so you can use natural-language feedback from your subject matter experts to improve the quality of the supervisor agent.

See Use Supervisor Agent to create a coordinated multi-agent system.

You can now use real-time mode, a trigger type for Structured Streaming that enables sub-second latency data processing. This mode is designed for operational workloads that require immediate response to streaming data. See Real-time mode concepts.

• Create a metric view from a dashboard dataset: Authors can now promote a dashboard dataset to a metric view using the AI/BI dashboard data tab. See Export as a metric view.

Databricks SQL version 2025.20 is rolling out in stages to the Current channel. For features and updates in this release, see 2025.20 features.

• Improvements to named parameters: Date-range and multi-select parameters are now supported. For date range parameters, see Add a date range using .min and .max. For multi-select parameters, see Pass multiple values as a string.

• Updated header layout in SQL editor: The run button and catalog picker have moved to the header, creating more vertical space for writing queries.

You can now use Databricks Git folders to track and manage changes to alerts. To track alerts with Git, place them in a Databricks Git folder. Newly cloned alerts only appear in the alerts list page or API after a user interacts with them. They have paused schedules and need to be explicitly resumed by users. See How Git integration works with alerts.

The Databricks documentation site now provides an RSS feed that contains updates to the product and other feature release notes. This feed can be used by any feed reader or client that can consume RSS, so that you can take advantage of feed reader features such as email notifications for Databricks product releases. See Databricks release notes feed.

You can now ask natural language questions about the sample data using Catalog Explorer. The Assistant generates the SQL based on metadata context and table usage patterns. After the query is generated, you can validate the query and then run it against the underlying table. See Explore table data using an LLM.

Databricks Runtime 17.1 and Databricks Runtime 17.1 ML are now in Beta. These releases include improvements to streaming, SQL functions, and connector behavior, along with reliability and performance enhancements across the platform.

See Databricks Runtime 17.1 (EoS) and Databricks Runtime 17.1 for Machine Learning (EoS).

Google Gemma 3 12B is now available on Mosaic AI Model Serving as a Databricks-hosted foundation model. See Databricks-hosted inference.

Gemma 3 12B model supports text inputs for the following Model Serving features:

• Foundation Model APIs pay-per-token.
• Foundation Model APIs provisioned throughput.
• AI Functions. Both real time inference and batch inference workloads are supported.

The CAN VIEW permission allows users to view SQL warehouses, including query history and query profiles. These users cannot run queries on the warehouse.

See SQL warehouse ACLs.

The simplified compute creation form is now enabled by default when you create all-purpose or jobs compute in the Databricks UI.

See Use the simple form to manage compute.