See SAP BDC semantic metadata.

This is a breaking change for users who paginate results using next_page_token. Tokens generated by a previous version of the API are not valid with the new version. If you use a token from a previous version, the API returns an error:

Invalid page_token: this token was generated by a previous/different API version. Please retry without page_token.

To continue paginating after this change, start a new request without a next_page_token.

• Databricks Runtime 18.1
• Databricks Runtime 18.0
• Databricks Runtime 17.3 LTS
• Databricks Runtime 16.4 LTS
• Databricks Runtime 15.4 LTS
• Databricks Runtime 14.3 LTS
• Databricks Runtime 13.3 LTS

See Deploy a Databricks app.

See Lakebase compliance.

Supervisor Agent is now available in the following additional Azure regions: uksouth and centralindia. Cross-geo processing must be enabled for these regions.

For the full list of supported regions, see Features with limited regional availability.

• Starting in late May 2026: The new SQL editor will be enabled by default for all workspaces. The ability to turn off the feature at the workspace level will no longer be available. Individual users will still be able to switch their queries to the legacy SQL editor after this period begins.
• Starting in late July 2026: The legacy SQL editor will be retired. All users will use the new SQL editor, and the individual opt-out will no longer be available.

To learn more about the new SQL editor, see Write queries and explore data in the new SQL editor. If you have questions about this transition, contact your account team.

• Classification: A visual UI for classifying documents into predefined categories. Configure and refine classification labels, preview results on sample documents, and then run classification at scale.

• Information Extraction: A visual UI for extracting structured data from documents, tables, PDFs, and images. Define and refine an extraction schema, preview results on sample documents, and then extract at scale.

Both experiences are built on the v2 signatures of ai_classify and ai_extract.

See ai_classify function and ai_extract function.

Observation metric errors no longer fail queries

Errors during observation metric collection no longer cause query execution failures. Previously, errors in OBSERVE clauses (such as division by zero) could block or fail the entire query. Now, the query completes successfully and the error is raised when you call observation.get.

FILTER clause for MEASURE aggregate functions

MEASURE aggregate functions now support FILTER clauses. Previously, filters were silently ignored.

Optimized writes for Unity Catalog CRTAS operations

CREATE OR REPLACE TABLE AS SELECT (CRTAS) operations on partitioned Unity Catalog tables now apply optimized writes by default, producing fewer, larger files. To disable, set spark.databricks.delta.optimizeWrite.UCTableCRTAS.enabled to false.

Timestamp partition values use session timezone

Timestamp partition values now use the SQL warehouse session timezone. If you have timestamp partitions written before Databricks SQL version 2025.40, run SHOW PARTITIONS to verify your partition metadata before writing new data.

DESCRIBE FLOW reserved keyword

The DESCRIBE FLOW command is now available. If you have a table named flow, use DESCRIBE schema.flow, DESCRIBE TABLE flow, or DESCRIBE `flow` with backticks.

SpatialSQL boolean set operations

ST_Difference, ST_Intersection, and ST_Union use a new implementation with the following improvements:

• Valid input geometries always produce a result and no longer raise errors. Invalid inputs don't raise errors but might not produce valid results.
• Approximately 2x faster performance.
• Results can differ after the 15th decimal place for line-segment intersections due to different formulas and order of operations.
• Results are normalized for consistent, comparable output:
  • Points are sorted by coordinate values.
  • Linestrings are built from the longest possible paths.
  • Polygon rings are rotated so the first point has the smallest coordinate values.
• This normalization applies in all cases except when calling ST_Difference with two non-overlapping geometries, where the first geometry is returned unmodified.

Exception types for SQLSTATE

Exception types now support SQLSTATE. If your code parses exceptions by string matching or catches specific exception types, update your error handling logic.

DATETIMEOFFSET data type support for Microsoft Azure Synapse

The DATETIMEOFFSET data type is now available for Microsoft Azure Synapse connections.

Google BigQuery table comments

Google BigQuery table descriptions are resolved and exposed as table comments.

Schema evolution with INSERT statements

Use the WITH SCHEMA EVOLUTION clause with SQL INSERT statements to automatically evolve the target table's schema during insert operations. The clause is supported for INSERT INTO, INSERT OVERWRITE, and INSERT INTO ... REPLACE forms. For example:

INSERT WITH SCHEMA EVOLUTION INTO students TABLE visiting_students_with_additional_id;

The target Delta Lake table's schema is updated to accommodate additional columns or widened types from the source. For details, see schema evolution and INSERT statement syntax.

Preserved NULL struct values in INSERT operations

INSERT operations with schema evolution or implicit casting now preserve NULL struct values when the source and target tables have differing struct field orders.

parse_timestamp SQL function

The parse_timestamp SQL function parses timestamp strings using multiple patterns and runs on the Photon engine for improved performance when parsing timestamps in multiple formats. See Datetime patterns for information about datetime pattern formatting.

max_by and min_by with optional limit

The aggregate functions max_by and min_by now accept an optional third argument limit (up to 100,000). When provided, the functions return an array of up to limit values corresponding to the largest (or smallest) values of the ordering expression, simplifying top-K and bottom-K queries without window functions or CTEs.

Vector aggregate and scalar functions

New SQL functions operate on ARRAY<FLOAT> vectors for embedding and similarity workloads:

Aggregate functions:

• vector_avg: Returns the element-wise average of vectors in a group.
• vector_sum: Returns the element-wise sum of vectors in a group.

Scalar functions:

• vector_cosine_similarity: Returns the cosine similarity of two vectors.
• vector_inner_product: Returns the inner (dot) product of two vectors.
• vector_l2_distance: Returns the Euclidean (L2) distance between two vectors.
• vector_norm: Returns the Lp norm of a vector (1, 2, or infinity).
• vector_normalize: Returns a vector normalized to unit length.

See Built-in functions.

SQL cursor support in compound statements

SQL scripting compound statements now support cursor processing. Use DECLARE CURSOR to define a cursor, then OPEN statement, FETCH statement, and CLOSE statement to run the query and consume rows one at a time. Cursors can use parameter markers and condition handlers such as NOT FOUND for row-by-row processing.

Approximate top-k sketch functions

New functions enable building and combining approximate top-K sketches for distributed top-K aggregation:

• approx_top_k_accumulate: Builds a sketch per group.
• approx_top_k_combine: Merges sketches.
• approx_top_k_estimate: Returns the top K items with estimated counts.

For more information, see approx_top_k aggregate function and Built-in functions.

Tuple sketch functions

New aggregate and scalar functions for tuple sketch support distinct counting and aggregation over key-summary pairs.

Aggregate functions:

• tuple_sketch_agg_double aggregate function
• tuple_sketch_agg_integer aggregate function
• tuple_union_agg_double aggregate function
• tuple_union_agg_integer aggregate function
• tuple_intersection_agg_double aggregate function
• tuple_intersection_agg_integer aggregate function

Scalar functions:

• tuple_sketch_estimate
• tuple_sketch_summary
• tuple_sketch_theta
• tuple_union
• tuple_intersection
• tuple_difference

See Built-in functions.

Custom dependencies for Unity Catalog Python UDTFs

Unity Catalog Python user-defined table functions (UDTFs) can now use custom dependencies for external libraries, so you can use packages beyond what's available in the default SQL warehouse environment. See Extend UDFs using custom dependencies.

New geospatial functions

The following geospatial functions are now available:

• st_estimatesrid function: Estimates the best projected spatial reference identifier (SRID) for an input geometry.
• st_force2d function: Converts a geography or geometry to its 2D representation.
• st_nrings function: Counts the total number of rings in a polygon or multipolygon, including both exterior and interior rings.
• st_numpoints function: Counts the number of non-empty points in a geography or geometry.

Photon support for geospatial functions

The following geospatial functions now run on the Photon engine for faster performance:

• st_difference function
• st_intersection function
• st_union function

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

See Databricks Apps.

For Azure China, the transition will be announced later.

The new URLs include the recipient ID in the domain:

https://<recipient-id>.delta-sharing.westus.azuredatabricks.net/api/2.0/delta-sharing/metastores/<metastore-id>

For reference, URLs created before this change do not contain the recipient ID.

https://westus.azuredatabricks.net/api/2.0/delta-sharing/metastores/<metastore-id>

The old URLs will continue to work for a period of time. The specific duration depends on the recipient type and the token creation date. Data providers should transition to the new URL format before the old URL format becomes invalid.

OIDC Federation sharing:

Data providers need to verify their recipients are using the new URL format before July 1, 2027. Starting July 1, 2026, providers can find the new URL in the Delta Sharing UI. After July 1, 2027, the old URL format will not be valid.

Bearer token sharing:

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

See AI Runtime.

• Upgrades the Simple Logging Facade for Java (SLF4J) from version 1.7.36 to 2.0.13 and moves slf4j-api to the root package path to align with standard SLF4J usage. The driver no longer depends on Log4j.
• The EnableTelemetry connection property is now on by default. The driver only sends telemetry when the workspace allows it.

This release also includes the following bug fixes:

• Fixes an incorrect error message when the driver can't write to the configured log path.
• Fixes a connectivity failure when using HTTP proxies that don't require authentication.

For more information, see Vector search cost management guide.

For more information on SQL warehouse sizing, see SQL warehouse sizing, scaling, and queuing behavior.

• OAuth roles: Create OAuth roles using the Lakebase UI or REST API, in addition to SQL. See Create Postgres roles.
• Budget policies and tags: Associate a serverless budget policy and add custom tags to your project for cost attribution and tracking. See Manage projects.

See Schedule refreshes in Databricks SQL.

• Databricks Runtime 18.0
• Databricks Runtime 17.3 LTS
• Databricks Runtime 17.2
• Databricks Runtime 16.4 LTS
• Databricks Runtime 15.4 LTS
• Databricks Runtime 14.3 LTS
• Databricks Runtime 13.3 LTS

See Transactions.

For a full list of supported compliance standards, see Compliance security profile.

To access this model, use:

• Foundation Model APIs pay-per-token
  • Query vision models
• Batch inference workloads using AI Functions

See Persist custom model serving data to Unity Catalog.

See Configure telemetry for Databricks Apps.

See Tree view of the workspace browser side panel (Public Preview).

See Lakebase Autoscaling limitations.

Use Knowledge Assistant to create a chatbot that can answer questions about your documents and provide high-quality responses with citations.

See Use Knowledge Assistant to create a high-quality chatbot over your documents.

See Install an external MCP server.

See High availability.

New regions:

• australiaeast (Australia East)
• brazilsouth (Brazil South)
• canadacentral (Canada Central)
• centralindia (Central India)
• southcentralus (South Central US)
• southeastasia (Southeast Asia)
• eastus (East US)
• centralus (Central US)
• northeurope (North Europe)
• uksouth (UK South)
• westus2 (West US 2)

See Get started with Lakebase Autoscaling.

February 27, 2026

This serverless compute release roughly corresponds to Databricks Runtime 18.0.

New features

• SQL scripting is now generally available: SQL scripting is now generally available.

• Redshift JDBC driver upgraded to 2.1.0.28: The Redshift JDBC driver has been upgraded to version 2.1.0.28.

• Shared isolation execution environment for Unity Catalog Python UDFs: Unity Catalog Python UDFs with the same owner can now share an isolation environment by default, improving performance and reducing memory usage. To ensure a UDF always runs in a fully isolated environment, add the STRICT ISOLATION characteristic clause. See Environment isolation.

• SQL window functions in metric views: You can now use SQL window functions in metric views to calculate running totals, rankings, and other window-based calculations.

• Dynamic shuffle partition adjustment in stateless streaming queries: You can now change the number of shuffle partitions in stateless streaming queries without restarting the query.

• Adaptive Query Execution and auto-optimized shuffle in stateless streaming queries: Adaptive Query Execution (AQE) and auto-optimized shuffle (AOS) are now supported in stateless streaming queries.

• Literal string coalescing everywhere: The ability to coalesce sequential string literals such as 'Hello' ' World' into 'Hello World' has been expanded to any place string literals are allowed. See STRING type.

• Parameter markers everywhere: You can now use named (:param) and unnamed (?) parameter markers virtually anywhere a literal value of the appropriate type can be used, including DDL statements and column types. See Parameter markers.

• IDENTIFIER clause everywhere: The IDENTIFIER clause, which casts strings to SQL object names, has been expanded to nearly everywhere an identifier is permitted, including column aliases and column definitions. See IDENTIFIER clause.

• New BITMAP_AND_AGG function: A new bitmap_and_agg aggregate function function is now available.

• New Theta sketch functions: A new library of functions for approximate distinct count and set operations using Datasketches Theta Sketch is now available, including theta_sketch_agg, theta_union_agg, theta_intersection_agg, and related functions.

• New KLL Sketch function library: A new library of functions for building KLL Sketches for approximate quantile computation is now available, including kll_sketch_agg_bigint, kll_sketch_agg_double, and related functions.

• Apache Parquet library upgraded to 1.16.0: The Apache Parquet library has been upgraded to version 1.16.0.

• New geospatial functions: The following new geospatial functions are now supported:

  • st_azimuth function: Returns the north-based azimuth from the first point to the second in radians.
  • st_boundary function: Returns the boundary of the input geometry.
  • st_closestpoint function: Returns the 2D projection of a point on the first geometry closest to the second geometry.
  • st_geogfromewkt function: Parses an Extended Well-Known Text (EWKT) description of a geography.
  • st_geomfromewkt function: Parses an Extended Well-Known Text (EWKT) description of a geometry.

• Improved spatial join performance: Spatial join performance is now improved by introducing shuffled spatial join support.

• Improved performance for geospatial functions: Photon implementations are now available for st_isvalid function, st_makeline function, and st_makepolygon function.

• EWKT input support: The try_to_geography, try_to_geometry, to_geography, and to_geometry functions now accept Extended Well-Known Text (EWKT) as input.

Behavior changes

• FSCK REPAIR TABLE includes metadata repair by default: The FSCK REPAIR TABLE command now includes an initial metadata repair step before checking for missing data files. The command can work on tables with corrupt checkpoints or invalid partition values.

• Proration factors aligned between reads and auto-optimized writes: Proration factors for partition sizing now use fractional values consistently across read operations and auto-optimized writes. This change might result in a different number of tasks for read operations.

• Python UDF execution unified across PySpark and Unity Catalog: Unity Catalog Python UDFs now use Apache Arrow as the default interchange format, improving overall performance. As part of this change, TIMESTAMP values passed to Python UDFs no longer include timezone information in the datetime object's tzinfo attribute. If your UDF relies on timezone information, restore it with date = date.replace(tzinfo=timezone.utc). See Timestamp timezone behavior for inputs.

• Improved error messages for Kafka connector login module issues: When using the Kafka connector with an unshaded login module class, Azure Databricks now provides error messages that suggest using the correct shaded class prefix (kafkashaded.org.apache.kafka or kafkashaded.software.amazon.msk.auth.iam).

• Time travel restrictions and VACUUM retention behavior: Azure Databricks now blocks time travel queries beyond the deletedFileRetentionDuration threshold for all tables. The VACUUM command ignores the retention duration argument except when the value is 0 hours. You can't set deletedFileRetentionDuration larger than logRetentionDuration or vice versa.

• BinaryType maps to bytes by default in PySpark: In PySpark, BinaryType now consistently maps to Python bytes. Previously, PySpark mapped BinaryType to either bytes or bytearray depending on the context. To restore the old behavior, set spark.sql.execution.pyspark.binaryAsBytes to false.

• Partition columns materialized in Parquet files: Partitioned Delta tables now materialize partition columns in newly written Parquet data files. Previously, partition values were stored in the Delta transaction log metadata and reflected in directory paths, but not written as columns in the Parquet files themselves. This change might affect workloads that directly read Parquet files written by Delta.

• DESCRIBE TABLE output includes metadata column: The output of DESCRIBE TABLE [EXTENDED] now includes a new metadata column for all table types, containing semantic metadata (display name, format, and synonyms) defined on the table as a JSON string.

• Databricks Runtime 18.0
• Databricks Runtime 17.3 LTS
• Databricks Runtime 17.2
• Databricks Runtime 16.4 LTS
• Databricks Runtime 15.4 LTS
• Databricks Runtime 14.3 LTS
• Databricks Runtime 13.3 LTS
• Databricks Runtime 12.2 LTS

See COMMENT ON.

• Databricks Runtime 18.0
• Databricks Runtime 17.3 LTS
• Databricks Runtime 17.2
• Databricks Runtime 16.4 LTS
• Databricks Runtime 15.4 LTS
• Databricks Runtime 14.3 LTS
• Databricks Runtime 13.3 LTS
• Databricks Runtime 12.2 LTS

• Domains: Business-aligned organization layer that groups data assets by functional area (for example, Marketing, or Finance) to improve discovery and stewardship
• Custom curation: Curators can customize and highlight specific assets in their organization or domain
• AI-powered recommendations: Surface popular and valuable datasets to enable curation at scale
• Unified discovery: Access tables, dashboards, Genie spaces, and more in one place

Domains are built on governed tags, allowing you to organize data assets in a way that's best suited for your business needs.

See Databricks Runtime 18.1 and Databricks Runtime 18.1 for Machine Learning.

See Enable OneLake catalog federation.

See AI Gateway (Beta).

When you create an external location using the UI, Azure Databricks checks for the required permissions and guides you to fix any issues or proceed with a force create. This check does not run in the API, but ingestion pipelines automatically fall back to prevent breakage.

You can disable file events for an external location after creation if needed. For more information, see Set up file events for an external location.

SQL scripting is generally available

SQL scripting is now generally available. Write procedural logic with SQL, including conditional statements, loops, local variables, and exception handling.

Parameter markers now supported in more SQL contexts

You can now use named (:param) and unnamed (?) parameter markers anywhere a literal value of the appropriate type is allowed. This includes DDL statements such as CREATE VIEW v AS SELECT ? AS c1, column types such as DECIMAL(:p, :s), and comments such as COMMENT ON t IS :comment. This allows you to parameterize a large variety of SQL statements without exposing your code to SQL injection attacks. See Parameter markers.

IDENTIFIER clause expanded to more SQL contexts

The IDENTIFIER clause, which casts strings to SQL object names, is now supported in nearly every context where an identifier is permitted. Combined with expanded parameter marker and literal string coalescing support, you can parameterize anything from column aliases (AS IDENTIFIER(:name)) to column definitions (IDENTIFIER(:pk) BIGINT NOT NULL). See IDENTIFIER clause.

Literal string coalescing supported everywhere

Sequential string literals such as 'Hello' ' World' now coalesce into 'Hello World' in any context where string literals are allowed, including COMMENT 'This' ' is a ' 'comment'. See STRING type.

New BITMAP_AND_AGG function

A new BITMAP_AND_AGG function is now available to complement the existing library of BITMAP functions.

New Theta Sketch functions for approximate distinct counts

A new library of functions for approximate distinct count and set operations using Datasketches Theta Sketch is now available:

• theta_sketch_agg aggregate function
• theta_union_agg aggregate function
• theta_intersection_agg aggregate function
• theta_sketch_estimate function
• theta_union function
• theta_difference function
• theta_intersection function

New KLL Sketch functions for approximate quantiles

A new library of functions for building KLL Sketches for approximate quantile computation is now available:

• kll_sketch_agg_bigint aggregate function
• kll_sketch_get_quantile_bigint function
• kll_sketch_merge_bigint function
• kll_sketch_agg_double aggregate function
• kll_sketch_get_quantile_double function
• kll_sketch_merge_double function
• kll_sketch_agg_float aggregate function
• kll_sketch_get_quantile_float function
• kll_sketch_merge_float function
• kll_sketch_get_n_bigint function
• kll_sketch_get_rank_bigint function
• kll_sketch_to_string_bigint function
• kll_sketch_get_n_double function
• kll_sketch_get_rank_double function
• kll_sketch_to_string_double function
• kll_sketch_get_n_float function
• kll_sketch_get_rank_float function
• kll_sketch_to_string_float function

You can merge multiple KLL sketches in an aggregation context using kll_merge_agg_bigint, kll_merge_agg_double, and kll_merge_agg_float.

SQL window functions in metric views

You can now use SQL window functions in metric views to calculate running totals, rankings, and other window-based calculations.

New geospatial functions

The following new geospatial functions are now available:

• st_azimuth function: Returns the north-based azimuth from the first point to the second in radians in [0, 2π).
• st_boundary function: Returns the boundary of the input geometry.
• st_closestpoint function: Returns the 2D projection of a point on the first geometry that is closest to the second geometry.
• st_geogfromewkt function: Parses an Extended Well-Known Text (EWKT) description of a geography.
• st_geomfromewkt function: Parses an Extended Well-Known Text (EWKT) description of a geometry.

EWKT input support for existing geometry and geography functions

The following functions now accept Extended Well-Known Text (EWKT) as input:

• try_to_geography function
• try_to_geometry function
• to_geography function
• to_geometry function

Improved performance for repeated queries over tables with row filters and column masks

Repeated eligible queries over tables with row filters and column masks now benefit from improved query result caching, resulting in faster execution times.

Improved geospatial function performance

Spatial join performance is improved with shuffled spatial join support. The following ST functions now have Photon implementations:

• st_isvalid function
• st_makeline function
• st_makepolygon function

FSCK REPAIR TABLE includes metadata repair by default

FSCK REPAIR TABLE now includes an initial metadata repair step before checking for missing data files, allowing it to work on tables with corrupt checkpoints or invalid partition values. Additionally, the dataFilePath column in the FSCK REPAIR TABLE DRY RUN output schema is now nullable to support new issue types where the data file path is not applicable.

DESCRIBE TABLE output includes metadata column

The output of DESCRIBE TABLE [EXTENDED] now includes a metadata column for all table types. This column contains semantic metadata (display name, format, and synonyms) defined on the table as a JSON string.

NULL structs preserved in MERGE, UPDATE, and streaming write operations

NULL structs are now preserved as NULL in Delta Lake MERGE, UPDATE, and streaming write operations that include struct type casts. Previously, NULL structs were expanded to structs with all fields set to NULL.

Partition columns materialized in Parquet files

Partitioned Delta Lake tables now materialize partition columns in newly written Parquet data files. Previously, partition values were stored only in the Delta Lake transaction log metadata. Workloads that directly read Parquet files written by Delta Lake sees additional partition columns in newly written files.

Timestamp partition values respect session timezone

Timestamp partition values are now correctly adjusted using the spark.sql.session.timeZone configuration. Previously, they were incorrectly converted to UTC using the JVM timezone.

Time travel restrictions updated

Azure Databricks now blocks time travel queries beyond the deletedFileRetentionDuration threshold for all tables. The VACUUM command ignores the retention duration argument except when the value is 0 hours. You cannot set deletedFileRetentionDuration larger than logRetentionDuration.

SHOW TABLES DROPPED respects LIMIT clause

SHOW TABLES DROPPED now correctly respects the LIMIT clause.

• HITRUST
• PCI-DSS
• TISAX
• UK Cyber Essentials Plus

For a list of compliance standards by region, see Compliance security profile.

See Configure default Python package repositories for more details.

See Using a provided queue for S3.

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.

Supervisor Agent is now Generally Available in select US regions for workspaces without Enhanced Security and Compliance features. For workspaces that have the compliance security profile enabled and HIPAA controls selected, Supervisor Agent will be Generally Available soon. See What's coming.

See ALTER STREAMING TABLE and ALTER MATERIALIZED VIEW. For general information about using ALTER with Lakeflow Spark Declarative Pipelines, see Use ALTER statements with pipeline datasets.

To access this model, use:

• Foundation Model APIs pay-per-token
  • Query reasoning models
  • Query vision models
• Batch inference workloads using AI Functions

Impact on user experience:

• Most AI features continue to work as expected with no action required.
• Some AI capabilities that depend on models not currently hosted in your Azure Geography might no longer be available with cross-geo processing disabled. See Availability of Designated Services in each Geo for more information on which features require cross-geo processing.
• If you rely on a feature that is no longer available in your Azure Geography, enable cross-geo processing to allow that feature to continue using models hosted outside your Azure Geography.
• For workspaces in these Azure Geographies, AI features will only use Databricks-hosted models if both partner-powered AI features and Enforce data processing within workspace Geography for Designated Services settings are disabled. If partner-powered AI features is disabled and Enforce data processing within workspace Geography for Designated Services is enabled, then AI assistive features are not be available in the workspace.

Impact on user experience:

• Most AI features continue to work as expected with no action required.
• Some AI capabilities that depend on models not currently hosted in your Azure Geography might no longer be available with cross-geo processing disabled. See Availability of Designated Services in each Geo for more information on which features require cross-geo processing.
• If you rely on a feature that is no longer available in your Azure Geography, enable cross-geo processing to allow that feature to continue using models hosted outside your Azure Geography.
• For workspaces in these Azure Geographies, AI features will only use Databricks-hosted models if both partner-powered AI features and Enforce data processing within workspace Geography for Designated Services settings are disabled. If partner-powered AI features is disabled and Enforce data processing within workspace Geography for Designated Services is enabled, then AI assistive features are not be available in the workspace.

Supported Azure regions:

• eastus2 (East US 2)
• westeurope (West Europe)
• westus (West US)

See Get started with Lakebase Autoscaling.

See Connect Genie Code to MCP servers.

See Availability of Designated Services in each Geo for more information on which features require cross-geo processing.

See Availability of Designated Services in each Geo for more information on which features require cross-geo processing.

See Schedule refreshes in Databricks SQL for more information.

For a reference of these tables, see Jobs system table reference.

See Configure compute resources for a Databricks app.

App compute sizing is coming soon to workspaces with the compliance security profile enabled.

• Databricks Runtime 18.0
• Databricks Runtime 17.3 LTS
• Databricks Runtime 17.2
• Databricks Runtime 17.1
• Databricks Runtime 16.4 LTS
• Databricks Runtime 15.4 LTS
• Databricks Runtime 14.3 LTS
• Databricks Runtime 13.3 LTS
• Databricks Runtime 12.2 LTS

Users in these regions can now use Knowledge Assistant to create a production-grade AI agent that can answer questions about their documents and provide high-quality responses with citations.

See Use Knowledge Assistant to create a high-quality chatbot over your documents.

• Store unlimited traces in Delta tables for long-term retention and analysis
• Query trace data directly using SQL through a Databricks SQL warehouse
• Manage access control through Unity Catalog schema and table permissions
• Ensure compatibility with other OpenTelemetry clients and tools

See Store MLflow traces in Unity Catalog and Query MLflow traces using MLflow Databricks SQL.

For more information, see Manage serverless workspace base environments.

• Prompt matching (previously value sampling)
• Format assistance (previously example values)
• Entity matching (previously value dictionaries)

If you prefer to continue using ODBC for new connections, you will have two options:

1. When creating a connection, select default as the Implementation.
2. In Power BI Desktop, disable the ADBC feature preview, which will restore ODBC as the default for all new connections.

To learn more about ADBC, see Configure ADBC or ODBC driver for Power BI and the Databricks Driver in the ADBC repository.

Knowledge Assistant is now generally available in select US regions for workspaces without Enhanced Security and Compliance features.

This feature is available for the Google Analytics, Salesforce, and ServiceNow connectors.

See Select rows to ingest.

• Databricks Runtime 17.3 LTS
• Databricks Runtime 17.2
• Databricks Runtime 17.1
• Databricks Runtime 16.4 LTS
• Databricks Runtime 15.4 LTS
• Databricks Runtime 14.3 LTS
• Databricks Runtime 13.3 LTS
• Databricks Runtime 12.2 LTS

See Clock-hour-aligned slicing logic.

See Extend Genie Code with agent skills.

• The agent can automate multiple steps. From a single prompt, it can retrieve relevant assets, generate and run code, fix errors automatically, and visualize results. It adds the ability to sample data and cell outputs to provide better results.
• The Assistant in Agent mode will choose between Azure OpenAI or Anthropic on Databricks (uses endpoints hosted by Databricks Inc. in AWS within the Databricks security perimeter), and is only available when the partner-powered AI features setting is enabled.
• Admins can disable the preview if needed until the feature reaches General Availability.

See Use Genie Code for data science, the blog post, and Partner-powered AI features.

This behavior enforces the Disable legacy features account setting available in existing Azure Databricks accounts. See Disable access to legacy features in new workspaces.

Contact your Databricks account team to request access to the preview.

For more details, see View last successful run and update notebook.

See Lakeflow Jobs.

See Context-based ingress control.

If you use Databricks on Azure Government, contact your account team to receive access to the driver.

Simplified navigation:

The duplicative Catalogs tab is removed to reduce redundancy and focus on a single catalog navigation surface. DBFS and Send feedback actions move into the kebab menu for a cleaner layout.

New Suggested section:

A new Suggested tab on the Catalog Explorer landing page highlights frequently used objects, example objects for first-time users, and user favorites. This helps you quickly re-engage with important assets or discover helpful starting points.

Consolidated entry points:

Related capabilities are grouped under clearer categories to reduce visual noise and improve findability:

• Govern – Entry point for governed tags, metastore administration, and data classification
• Connect – Entry points for external locations, external data, credentials, and connections
• Share – Entry points for Delta Sharing and Clean Rooms

These groupings replace scattered sub-tabs and create a more intuitive, scalable information architecture.

• Disable legacy features: Account-level setting that disables access to DBFS, Hive Metastore, and No-isolation shared compute in new workspaces.
• Disable access to Hive metastore: Workspace-level setting that disables access to the Hive metastore used by your workspace.

For an in-depth comparison of the SharePoint connectors, see Choose your SharePoint connector.

To start using Auto Loader with file events, see the following:

• (Prerequisite) Enable file events for an external location
• File notification mode with and without file events enabled on external locations

See Use ForEachBatch to write to arbitrary data sinks in pipelines.

See Databricks Runtime 18.0 and Databricks Runtime 18.0 for Machine Learning.

• Databricks Runtime 17.3 LTS
• Databricks Runtime 17.2
• Databricks Runtime 17.1
• Databricks Runtime 17.0
• Databricks Runtime 16.4 LTS
• Databricks Runtime 15.4 LTS
• Databricks Runtime 14.3 LTS
• Databricks Runtime 13.3 LTS
• Databricks Runtime 12.2 LTS

jobs table: trigger, trigger_type, run_as_user_name, creator_user_name, paused, timeout_seconds, health_rules, deployment, create_time

job_tasks table: timeout_seconds, health_rules

job_run_timeline table: source_task_run_id, root_task_run_id, compute, termination_type, setup_duration_seconds, queue_duration_seconds, run_duration_seconds, cleanup_duration_seconds, execution_duration_seconds

job_task_run_timeline table: compute, termination_type, task_parameters, setup_duration_seconds, cleanup_duration_seconds, execution_duration_seconds

pipelines table: create_time

These columns are not populated for rows emitted before early December 2025. See Jobs system table reference.

Existing open sharing recipient tokens issued before December 8, 2025, with expiration dates after December 8, 2026, or with no expiration date, automatically expire on December 8, 2026. If you currently use recipient tokens with long or unlimited lifetimes, review your integrations and renew tokens as needed to avoid breaking changes after this date.

See Create a recipient object for non-Databricks users using bearer tokens (open sharing).

To access this model, use:

• Foundation Model APIs pay-per-token
  • Query reasoning models
  • Query vision models
• Batch inference workloads using AI Functions

• australiasoutheast
• canadacentral
• mexicocentral
• norwayeast
• southafricanorth

• Databricks Runtime 17.3 LTS
• Databricks Runtime 17.2
• Databricks Runtime 17.1
• Databricks Runtime 17.0
• Databricks Runtime 16.4 LTS
• Databricks Runtime 15.4 LTS
• Databricks Runtime 14.3 LTS
• Databricks Runtime 13.3 LTS
• Databricks Runtime 12.2 LTS

See Manage network policies for serverless egress control.

See Add tables and schemas secured by ABAC policies to a share and Read ABAC-secured data and apply ABAC policies.

Minimum Databricks Runtime versions:

• ALTER TABLE SET MANAGED: DBR 17.3 LTS or above
• ALTER TABLE SET EXTERNAL: DBR 17.0 LTS or above
• ALTER CATALOG DROP CONNECTION: DBR 17.3 LTS or above
• ALTER SCHEMA SET MANAGED LOCATION: DBR 16.4 LTS or above

• Governed access to data sources using Unity Catalog connections
• Create the connection one time and reuse it across any Unity Catalog compute type: serverless, standard clusters, dedicated clusters, and Databricks SQL
• Stable for Spark and compute upgrades
• Connection credentials are hidden from the querying user

The JDBC connection is available on Databricks Runtime 17.3 and above on standard or dedicated access mode, or on serverless compute. For more information, see JDBC connection.

See Share data between SAP Business Data Cloud (BDC) and Azure Databricks.

See MCP server listing fields and Get access to external MCP servers.

See Model Context Protocol (MCP) on Databricks.

• Databricks Runtime 17.3 LTS
• Databricks Runtime 17.2
• Databricks Runtime 17.1
• Databricks Runtime 17.0
• Databricks Runtime 16.4 LTS
• Databricks Runtime 15.4 LTS
• Databricks Runtime 14.3 LTS
• Databricks Runtime 13.3 LTS
• Databricks Runtime 12.2 LTS

• See Function calling on Azure Databricks
• See Structured outputs on Azure Databricks

Qwen3-Next 80B A3B Instruct is available in Beta. See Manage Azure Databricks previews for guidance on how to enable this model based on your Databricks plan tier.

In Public Preview, ABAC includes two key changes from the Beta:

• Account-level enforcement: ABAC has transitioned from a workspace-level preview to an account-level preview. ABAC policies are now enforced across all workspaces attached to an account’s metastore by default. See ABAC Beta to Public Preview transition.

• Automatic type casting for column masks: Databricks now automatically casts the output of column mask functions defined in ABAC policies to match the target column’s data type. This ensures consistent typing and more reliable query behavior. Existing column mask functions might fail if their return types are incompatible with the target column’s type. See ABAC column mask type casting

For more information, see Unity Catalog attribute-based access control (ABAC).

See Step into workspace files.

EXECUTE IMMEDIATE using constant expressions

You can now pass constant expressions as the SQL string and as arguments to parameter markers in EXECUTE IMMEDIATE statements.

LIMIT ALL support for recursive CTEs

You can now use LIMIT ALL to remove the total size restriction on recursive common table expressions (CTEs).

st_dump function support

You can now use the st_dump function to get an array containing the single geometries of the input geometry. See st_dump function.

Polygon interior ring functions are now supported

You can now use the following functions to work with polygon interior rings:

• st_numinteriorrings: Get the number of inner boundaries (rings) of a polygon. See st_numinteriorrings function.
• st_interiorringn: Extract the n-th inner boundary of a polygon and return it as a linestring. See st_interiorringn function.

Support MV/ST refresh information in DESCRIBE EXTENDED AS JSON

Azure Databricks now generates a section for materialized view and streaming table refresh information in the DESCRIBE EXTENDED AS JSON output, including last refreshed time, refresh type, status, and schedule.

Add metadata column to DESCRIBE QUERY and DESCRIBE TABLE

Azure Databricks now includes a metadata column in the output of DESCRIBE QUERY and DESCRIBE TABLE for semantic metadata.

For DESCRIBE QUERY, when describing a query with metric views, semantic metadata propagates through the query if dimensions are directly referenced and measures use the MEASURE() function.

For DESCRIBE TABLE, the metadata column appears only for metric views, not other table types.

Correct handling of null structs when dropping NullType columns

When writing to Delta tables, Azure Databricks now correctly preserves null struct values when dropping NullType columns from the schema. Previously, null structs were incorrectly replaced with non-null struct values where all fields were set to null.

See Create and manage compute policies and Compute policy reference.

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

New features

The connector now supports Kerberos with proxy connections. To enable Kerberos proxy, set UseProxy=1 and ProxyAuth=2. To set proxy details, use ProxyHost, ProxyPort, ProxyKrbRealm, ProxyKrbFQDN, and ProxyKrbService.

Resolved issues

• Fixed an issue where the connector failed to run complex queries that contained ? characters in native mode.
• Fixed intermittent failures in Unity Catalog volume ingestion caused by unexpected connector behavior.
• Fixed an assertion error in getColumns when a table included a column of type Void or Variant and the java -ea flag was enabled.

See Compatibility Mode.

Existing tables continue to use their current compression codec. To change the compression codec for an existing table, set the delta.parquet.compression.codec table property. See Table properties reference.

See What changes are in the Unified Runs List preview?.

See Databricks Connector for Google Sheets.

See Manage dashboard tags and Add tags.

See Collaborate on bundles in the workspace.

See Share data between SAP Business Data Cloud (BDC) and Azure Databricks.

This multimodal support is available using the following functionalities:

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

See Context-based ingress control.

For billing log reference, see Billable usage system table reference.

See Configure Databricks Serverless Private Git

See Databricks Runtime 17.3 LTS and Databricks Runtime 17.3 LTS for Machine Learning.

For details, see Apply row filters and column masks.

• Thinking messages content in the messages.content array.
• Images content blocks in the messages.content array.
• Tool use, results and definitions in the tools array.

See Foundation model REST API reference.

• The cell execution minimap now appears in the right margin of notebooks. Use the minimap to get a visual overview of your notebook's run status and quickly navigate between cells. See Cell execution minimap.
• Use Databricks Assistant to help diagnose and fix environment errors, including library installation errors. See Debug environment errors.
• When reconnecting to serverless notebooks, sessions are automatically restored with the notebook's Python variables and Spark state. See Automated session restoration for serverless notebooks.
• Pyspark authoring completion now supports agg, withColumns, withColumnsRenamed, and filter/where clauses. See Get inline code suggestions: Python and SQL examples.
• Databricks now supports importing and exporting IPYNB notebooks up to 100 MB. Revision snapshot autosaving, manual saving, and cloning are supported for all notebooks up to 100 MB. See Notebook sizing.
• When cloning and exporting notebooks, you can now choose whether to include cell outputs or not. See Manage notebook format.

• See Enrich data using AI Functions for an example.
• See Supported foundation models on Mosaic AI Model Serving for region availability.

To use semantic metadata, your metric view must use YAML specification version 1.1 or higher and run on Databricks Runtime 17.2 or above. The corresponding Databricks SQL version is 2025.30, available in the Preview channel for SQL warehouses.

See Use semantic metadata in metric views and Upgrade considerations.

To use semantic metadata, your metric view must use YAML specification version 1.1 or higher and run on Databricks Runtime 17.2 or above. The corresponding Databricks SQL version is 2025.30, available in the Preview channel for SQL warehouses.

See Use semantic metadata in metric views and Upgrade your YAML to 1.1.

The email you provide becomes the GIT_AUTHOR_EMAIL and GIT_COMMITTER_EMAIL for commits, allowing Git providers to properly associate commits with your user account and display your profile information. If no email is specified, Databricks uses your Git username as the email address (legacy behavior).

See Git commit identity and email configuration.

This is a legitimate request from Databricks. It asks you to approve a new permission that allows Azure Databricks to read your GitHub account email(s). Granting this permission will let Azure Databricks retrieve and save your primary GitHub account email to your Linked Git credential in Azure Databricks. In an upcoming feature, this will ensure that commits made from Azure Databricks are properly linked to your GitHub identity.

If you don't accept the new permission, your Linked Git credential will still authenticate with GitHub. However, future commits from this credential will not be associated with your GitHub account identity

See Pipeline update timeline table schema.

See Develop and debug ETL pipelines with the Lakeflow Pipelines Editor.

For background, see Prepare for Connected App Usage Restrictions Change in the Salesforce documentation.

For steps to connect to Salesforce, see Salesforce

UTF8 based collations now support LIKE operator

You can now use LIKE with columns that have one of the following collations enabled: UTF8_Binary, UTF8_Binary_RTRIM, UTF8_LCASE, UTF8_LCASE_RTRIM. See Collation.

ST_ExteriorRing function is now supported

You can now use the ST_ExteriorRing function to extract the outer boundary of a polygon and return it as a linestring. See st_exteriorring function.

Declare multiple session or local variables in a single DECLARE statement

You can now declare multiple session or local variables of the same type and default value in a single DECLARE statement. See DECLARE VARIABLE and BEGIN END compound statement.

Support TEMPORARY keyword for metric view creation

You can now use the TEMPORARY keyword when creating a metric view. Temporary metric views are visible only in the session that created them and are dropped when the session ends. See CREATE VIEW.

DESCRIBE CONNECTION shows environment settings for JDBC connections

Azure Databricks now includes user-defined environment settings in the DESCRIBE CONNECTION output for JDBC connections that support custom drivers and run in isolation. Other connection types remain unchanged.

SQL syntax for Delta read options in streaming queries

You can now specify Delta read options for SQL-based streaming queries using the WITH clause. For example:

SELECT * FROM STREAM tbl WITH (SKIPCHANGECOMMITS=true, STARTINGVERSION=X);

Correct results for split with empty regex and positive limit

Azure Databricks now returns correct results when using split function with an empty regex and a positive limit. Previously, the function incorrectly truncated the remaining string instead of including it in the last element.

Fix url_decode and try_url_decode error handling in Photon

In Photon, try_url_decode() and url_decode() with failOnError = false now return NULL for invalid URL-encoded strings instead of failing the query.

See Enable the default publishing mode in a pipeline.

See Automatic authentication passthrough.

See Query route-optimized serving endpoints.

With Databricks One, business users can:

• View and interact with AI/BI dashboards to track KPIs and analyze metrics.
• Ask data questions in natural language using Genie.
• Use custom-built Databricks Apps that combine analytics, AI, and workflows.

Workspace admins can enable Databricks One from the Previews page in the admin console.

See What is Databricks One?.

After Delta Sharing for Default Storage – Expanded Access is enabled by default for all Azure Databricks users, Delta Sharing on Lakehouse Federation will automatically be available in regions where default storage is supported.

See Default storage in Databricks and Add foreign schemas or tables to a share.

• (Pre-requisite) Enable file events for an external location
• File notification mode with and without file events enabled on external locations

For information about Python custom data sources, see the following:

• Load data from a Python custom data source.
• Create a sink.
• PySpark custom data sources.

See Sync users and groups automatically from Microsoft Entra ID.

See Databricks Online Feature Stores.

For information, see MLflow system tables reference.

The Data Science Agent is custom-built for data science workflows and can build an entire notebook for tasks like EDA, forecasting, and machine learning from scratch. Using your prompt, it can plan a solution, retrieve relevant assets, run code, use cell outputs to improve results, fix errors automatically, and more.

See Use Genie Code for data science.

See Limitations.

See Limitations.

See Enable the default publishing mode in a pipeline

See Use your Azure Databricks data to build Power Automate flows.

See Create and manage a database instance.

• Calling GET with filter params filter=displayName eq value_without_quotes results in a syntax error. To prevent this error, use quotation marks to wrap the value (for example, filter=displayName eq "value_with_quotes").

• Calling GET /api/2.0/accounts/{account_id}/scim/v2/Groups no longer returns members. Instead, iterate through get group details to get membership information. See get group details.

• Calling PATCH /api/2.0/accounts/{account_id}/scim/v2/Groups/{id} returns a 204 response instead of a 200 response.

These changes apply to new accounts and will be grandfathered into existing accounts. Current integrations will continue to work without disruption.

For information about using the cell execution minimap, see Navigate the Databricks notebook and file editor.

Workspace admins can now create and manage the base environments available in their workspace. They can also set a default base environment for all new serverless notebooks. This allows workspace users to quickly start working in a consistent, cached environment.

See Manage serverless workspace base environments.

See Supported source tables.

See Model Context Protocol (MCP) on Databricks.

See Enable the default publishing mode in a pipeline

See Governed tags.

See Databricks Runtime 17.2 (EoS) and Databricks Runtime 17.2 for Machine Learning (EoS).

See Canada Protected B

See Configure a federation policy.

See Data type mapping.

See Automatic liquid clustering, create_streaming_table, table, CREATE MATERIALIZED VIEW (pipelines), and CREATE STREAMING TABLE (pipelines).

See Update a serverless budget policy of a database instance, Create a synced table, and Custom tags.

• germanywestcentral
• brazilsouth
• switzerlandnorth
• centralindia
• japaneast

See Features with limited regional availability

See Lakehouse Federation for Salesforce Data 360 File Sharing.

• Migrate a workspace from a Azure Databricks-managed VNet to your own VNet (VNet injection).
• Move a VNet injection workspace to a new VNet.
• Replace existing subnets in a VNet injection workspace.

See Update workspace network configuration.

See Personalized autocomplete.

When you deploy a bundle with a database instance, the instance starts running immediately.

See database_catalog, database_instance, and synced_database_table.

To learn more about chatting with the Assitant, see Tour of Genie Code pane. To learn more about compute resources, see Compute.

For information about setting the run-as user, see Set the run-as user.

Recursive common table expressions (rCTE) are generally available

Recursive common table expressions (rCTEs) are generally available. Navigate hierarchical data using a self-referencing CTE with UNION ALL to follow the recursive relationship.

Support for schema and catalog level default collation

You can now set a default collation for schemas and catalogs. This allows you to define a collation that applies to all objects created within the schema or catalog, ensuring consistent collation behavior across your data.

Support for Spatial SQL expressions and GEOMETRY and GEOGRAPHY data types

You can now store geospatial data in built-in GEOMETRY and GEOGRAPHY columns for improved performance of spatial queries. This release adds more than 80 new spatial SQL expressions, including functions for importing, exporting, measuring, constructing, editing, validating, transforming, and determining topological relationships with spatial joins. See ST geospatial functions, GEOGRAPHY type, and GEOMETRY type.

Support for schema and catalog level default collation

You can now set a default collation for schemas and catalogs. This allows you to define a collation that applies to all objects created within the schema or catalog, ensuring consistent collation behavior across your data.

Better handling of JSON options with VARIANT

The from_json and to_json functions now correctly apply JSON options when working with top-level VARIANT schemas. This ensures consistent behavior with other supported data types.

Support for TIMESTAMP WITHOUT TIME ZONE syntax

You can now specify TIMESTAMP WITHOUT TIME ZONE instead of TIMESTAMP_NTZ. This change improves compatibility with the SQL Standard.

Resolved subquery correlation issue

Azure Databricks no longer incorrectly correlates semantically equal aggregate expressions between a subquery and its outer query. Previously, this could lead to incorrect query results.

Error thrown for invalid CHECK constraints

Azure Databricks now throws an AnalysisException if a CHECK constraint expression cannot be resolved during constraint validation.

Stricter rules for stream-stream joins in append mode

Azure Databricks now disallows streaming queries in append mode that use a stream-stream join followed by window aggregation, unless watermarks are defined on both sides. Queries without proper watermarks can produce non-final results, violating append mode guarantees.

• Materialized views and streaming tables created after August 14, 2025 will have the warehouse timeout automatically applied.
• For materialized views and streaming tables created before August 14, 2025, run CREATE OR REFRESH to synchronize the timeout setting with the timeout configuration of the warehouse.
• All materialized views and streaming tables now have a default timeout of two days.

See New ETL pipeline.

See Azure Databricks and Microsoft Power Platform integration.

See Configure AI Gateway on model serving endpoints for how to configure token-based rate limits.

See Databricks-hosted models for AI-powered features.

See Function calling on Azure Databricks.

Admins can also add workspace-wide instructions, which take priority over user instructions.

Instructions apply across Assistant experiences like Inline Assistant, General Chat, Suggest Fix, and Edit mode. See Add custom instructions.

See Compute configuration reference.

This change affects SQL commands ([CREATE OR] REPLACE TABLE), DataFrame APIs (saveAsTable, replace, createOrReplace), and other similar update table operations.

See Manually apply row filters and column masks.

Users can request access to Unity Catalog objects that they discover. These requests are sent to configured destinations, such as emails, Slack or Microsoft Teams channels, or they can be redirected to an internal access management system.

You can also enable default email destinations so requests automatically go to the catalog or object owner’s email if no other destination is set. This ensures that access requests are delivered even when no destination is manually configured for an object.

See Manage access requests.

See Disable access to legacy features in new workspaces.

See Create and manage a database instance and Manage Azure Databricks previews.

See Run jobs continuously.

See Create external Delta tables from external clients.

This release includes the following fixes and new features:

• The process name is now used as the default UserAgentEntry if the UserAgentEntry is not explicitly set.
• Added support for Databricks domains cloud.databricks.us and cloud.databricks.mil.
• Enhanced recognition and handling of timestamp_ntz columns across multiple data source functions including SQLGetTypeInfo, SQLColumns, and SQLColAttribute.
• Added CRL (Certificate Revocation Lists) cache support on Windows when UseSystemTruststore is enabled.
• Added VOID type column support, so that VOID columns are now correctly listed in SQLGetColumns.
• Enabled OAuth Token exchange for IDPs different from the host, which allows the exchange of OAuth access tokens (including BYOT) for Databricks in-house tokens.
• Added support for Windows Server 2025.
• Fixed a memory leak in the driver.

This release includes upgrades to several third-party libraries:

• OpenSSL upgraded from 3.0.15 to 3.0.16
• libcURL upgraded from 8.11.0 to 8.12.1
• Expat upgraded from 2.6.3 to 2.7.1

This release includes the following behavior changes:

• The connector no longer supports Databricks Runtime version 10.4 LTS.
• The default maximum catalog name length and maximum schema name length has been changed from 128 to 1024.
• Ubuntu 20.04 is no longer supported.

For complete configuration information, see the Databricks ODBC Driver Guide installed with the driver download package.

See Configure service principals on Azure Databricks for Power BI.

For information about using Assistant on code, see Get coding help from Genie Code.

See the Terraform documentation and changelog.

See Billable usage system table reference and Serve lakehouse data with synced tables (Lakebase Provisioned).

The new form includes the following changes:

• New definition dropdown menus allow you to configure rules without needing to reference the policy syntax.
• Max compute resources per user, max DBUs per hour, and cluster type settings have moved under the Advanced options section.
• Tagging definitions now have their own separate section.
• Policy permission settings have moved out of the policy form and are now set using the permissions modal in the policy overview page.

See Configure policy definitions using UI elements.

See Add tables and schemas secured by ABAC policies to a share and Read ABAC-secured data and apply ABAC policies.

See Add streaming tables to a share, Add materialized views to a share, and Read shared streaming tables and materialized views.

See Enable Delta Sharing on a metastore and View providers.

• Add a split view to edit notebooks side by side. See Edit notebooks side by side.
• Pressing Cmd + F (Mac) or Ctrl + F (Windows) in a notebook now opens the native Databricks find-and-replace tool. This allows you to quickly search and replace text throughout your entire notebook, including content outside the current viewport. See Find and replace text.
• Quickly switch between tab groups based on authoring contexts using the , , and icons on the top left in the editor. See Switch between authoring contexts.

• centralus
• eastus
• eastus2
• northcentralus
• westus
• australiaeast
• brazilsouth
• canadacentral
• centralindia
• eastasia
• francecentral
• germanywestcentral
• japaneast
• koreacentral
• northeurope
• southeastasia
• swedencentral
• switzerlandnorth
• uaenorth
• uksouth
• westeurope

See ai_parse_document function.

• 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.

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.

See INSERT in the SQL language reference.

See Create Dataverse virtual tables with your Azure Databricks data.

See Add Azure Databricks to a data policy.

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

Azure does not support disabling public network access to OS disks. Because OS disks don't contain customer data, this doesn't pose a security risk.

See Data security and encryption.

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

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.

See SQL warehouse ACLs.

See Use the simple form to manage compute.

See Automated session restoration for serverless notebooks.

See How Git integration works with alerts.

See Configure ADBC or ODBC driver for Power BI.

See Add a Lakebase resource to a Databricks app.

See Resource limits.

See Configure Git integration for Git folders.

See IP addresses and domains for Azure Databricks services and assets.

Admins must add cross-cloud S3 locations to their network policy using the network policies API.

See Create a storage credential for connecting to AWS S3 (read-only).

See AI Runtime.

This external lineage metadata feature is in Public Preview. See Bring your own data lineage.

See Use PDFs for Information Extraction and Custom LLM.

This function is available in Beta. See ai_parse_document function.

See Databricks Runtime 17.0 (EoS) and Databricks Runtime 17.0 for Machine Learning (EoS).

See Use Open ID Connect (OIDC) federation to enable authentication to Delta Sharing shares (open sharing).

See Enable the default publishing mode in a pipeline.

• Endpoint: Specify the overall rate limit for all traffic passing through the endpoint, regardless of individual or user group limits.
• User (Default): Specify the rate limit for all users of the endpoint that do not have user-specific or group-specific rate limits defined.
• Specific user: Specify the rate limit for a particular user of the endpoint.
• Service principal: Specify the rate limit for a particular service principal.
• User group: Specify the rate limit for a particular user group. This limit is shared across all members of the group.

See Configure AI Gateway using the UI.

• Customizable dashboard theme: Authors can now customize the dashboard theme by setting custom font, background, and widget colors. See Dashboard settings.

If your organization uses resource firewalls to control inbound access:

• If your firewall rules reference the Azure Databricks service tag, no action is required.
• If you allow specific control plane public IPs, you must add all the outbound control plane IPs by September 26, 2025.

The previous outbound control plane IPs continue to be supported.

• If you already use a Databricks-hosted model for Assistant, there is no change.
• If you turn off Partner-powered AI features, AI documentation now stays available and uses a Databricks-hosted model.

For more information, see:

• Partner-powered AI features
• Add AI-generated comments to Unity Catalog objects

Using ABAC, you can define scalable policies once and apply them across large sets of data assets. Policies inherit across the object hierarchy and can include row-level filters or column masking logic. This simplifies governance, supports centralized policy management, and improves security posture. See Unity Catalog attribute-based access control (ABAC).

• Lower latency: Dashboards, ETL jobs, and mixed workloads now run faster, with up to 25% improvement. The upgrade is automatically applied to serverless SQL warehouses with no additional cost or configuration.
• Predictive Query Execution (PQE): PQE monitors tasks in real time and dynamically adjusts query execution to help avoid skew, spills, and unnecessary work.
• Photon vectorized shuffle: Keeps data in compact columnar format, sorts it within the CPU's high-speed cache, and processes multiple values simultaneously using vectorized instructions. This improves throughput for CPU-bound workloads such as large joins and wide aggregation.

See Lakebase.

The Beta release supports API-based pipeline creation. UI-based pipeline creation is not yet supported.

AI Gateway brings governance, monitoring, and production readiness to model serving endpoints using the following features:

• Permission and rate limiting to control who has access and how much access.
• Payload logging to monitor and audit data being sent to model APIs using inference tables.
• Usage tracking to monitor operational usage on endpoints and associated costs using system tables.
• Traffic routing to load balance traffic across multiple models.
• Fallbacks for external models to minimize production outages during and after deployment.

AI Guardrails remains in Public Preview.

For information about the SQL AUTO CDC API, see:

• AUTO CDC INTO (pipelines)

For information about the Python create_auto_cdc_flow APIs, see

• create_auto_cdc_flow
• create_auto_cdc_from_snapshot_flow

MLflow 3.0 on Azure Databricks delivers state-of-the-art experiment tracking, observability, and performance evaluation for machine learning models, generative AI applications, and agents on the Databricks Lakehouse. See Get started with MLflow 3 for models.

Deployment jobs allow you to automate tasks like evaluation, approval, and deployment whenever a new model version is created, integrating seamlessly with Unity Catalog models and Lakeflow Jobs. See MLflow 3 deployment jobs.

When the Performance optimized setting is enabled, your workload is optimized for faster startup and execution time. When disabled, the serverless workload runs on standard performance mode, which is optimized for cost and has a slightly higher launch latency.

For more information, see Select a performance mode and Select a performance mode.

• Managed MCP servers: Use Databricks-hosted servers for easy, no-maintenance access to Unity Catalog data and tools.
• Custom MCP servers: Host your own MCP server or third-party server as a Databricks app.

See Model Context Protocol (MCP) on Databricks.