Databricks Release Notes

Supervisor Agent is now available by default for workspaces with the compliance security profile enabled and HIPAA, FedRAMP Moderate, PCI-DSS, or HITRUST controls selected.

Use Supervisor Agent to create a multi-agent supervisor system that orchestrates AI agents and tools to work together on complex tasks.

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

Type widening for Delta Lake tables is now generally available. Type widening allows you to change column data types to wider types such as INT to BIGINT or FLOAT to DOUBLE during ingestion without rewriting underlying data files.

See Type widening.

Auto Loader now supports automatic type widening with the new addNewColumnsWithTypeWidening schema evolution mode. This feature automatically widens data types (such as int to long or float to double) without requiring data rewrite or user intervention. Type widening is available for all file formats with schema evolution support in Auto Loader including text formats (JSON, CSV, XML) and binary formats (Avro, Parquet).

This feature is in Public Preview and requires DBR 16.4 or above. See Automatic type widening with Auto Loader.

The entity_metadata struct in the system.access.table_lineage and system.access.column_lineage system tables now includes genie_space_id and alert_id fields. The genie_space_id field is populated when the lineage record originates from a Genie Agent querying data, and alert_id is populated when the lineage record originates from an alert query. For the full schema, see Entity metadata reference.

The new SQL editor now supports a Dynamic dropdown widget type for named parameter markers. Select a saved query to populate dropdown values, keeping the list current as your data changes. See Dynamic dropdown.

You can now connect external data sources such as Google Drive or SharePoint to a chat conversation in Genie One, allowing chat to search your company documents when answering questions. Each user authenticates individually using their own credentials. You can also connect custom MCP servers as additional data sources. See Connect Genie One to external tools and sources.

Supervisor Agent now supports long-running task mode, which automatically breaks complex tasks into multiple request/response cycles using task continuation to avoid timeouts. See Supervisor Agent long-running tasks.

You can now create and view alerts for anomaly detection directly in the Databricks Data Quality Monitoring UI. This feature is in Beta. See Alerts for anomaly detection.

Designed for deep learning workloads, AI Runtime brings GPU support to Databricks serverless compute. AI Runtime for single-node tasks is now in Public Preview.

The distributed training API for multi-GPU workloads remain in Beta.

See AI Runtime.

You can now add a SQL alert task to your Databricks jobs to evaluate Databricks SQL alerts as part of your workflows. SQL alert tasks enable automated condition checks and notifications within your data pipelines. See SQL alert task for jobs.

The usage dashboard is now generally available. Account admins can import customizable, pre-built usage dashboards to monitor their Databricks usage at the account and workspace levels. See Usage dashboards.

You can now use uv to manage Python dependencies in your Databricks apps using pyproject.toml and uv.lock. See Define Python dependencies with uv and Deployment logic.

Searching for measures and dimensions in the metric view overview page now returns correct results across paginated results and supports filtering by display name. See Unity Catalog metric views.

Mosaic AI Model Serving now supports OpenAI GPT-5.4 mini and GPT-5.4 nano as Databricks-hosted models. You can access these models using Foundation Model APIs pay-per-token.

Customers are responsible for ensuring their compliance with the terms of OpenAI's Acceptable Use Policy.

Databricks Asset Bundles has been renamed to Declarative Automation Bundles. See Why was Databricks Asset Bundles renamed to Declarative Automation Bundles?.

You can now add Lakebase databases as resources for your Databricks app. See Add a Lakebase resource to a Databricks app.

ai_parse_document improves bounding box and layout detection and addresses some common reading order errors. See ai_parse_document function.

You can now view health indicators for tables on schema and table overview pages in Catalog Explorer (Public Preview). Health indicators show a summary of each table's freshness and completeness status based on anomaly detection. See Anomaly detection.