Databricks releases, every cloud
- OpenAI GPT 5 models are now generally available in Mosaic AI Model Serving
Includes GPT-5 mini and nano models.
- Gemini 2.5 Pro and Flash models now supported on pay-per-token endpoints
Supports pay-per-token endpoints and Function calling.
- Serverless compute has been updated to version 17.3
Updated from version 17.2, requires no configuration changes.
- Alibaba Cloud Qwen3-Next Instruct now available in Beta
Supports 80B A3B Instruct in us-west-2 and ap-northeast-1 regions via pay-per-token APIs.
- New alert editing experience
Replaces old editor with a multi-tab editor for unified workflow.
- Improved autoscaling behavior for Mosaic AI Model Serving
Ignores brief traffic surges, scaling on sustained load increases. Reduces unnecessary concurrency scaling.
- Run the change, do not just read it. Hands-on labs in your own Databricks workspace, graded when you submit. Browse labs
- SQL MCP server now available (Beta)
Executes SQL queries against Unity Catalog tables using SQL warehouses. Available at https://<workspace-hostname>/api/2.0/mcp/sql.
- The billable usage table now records the performance mode of serverless jobs and pipelines
Records performance mode in product_features.performance_target column with values PERFORMANCE_OPTIMIZED, STANDARD, or null.
- Visualizations fix
Fixes legend selection for charts with aliased series names in SQL editor and notebooks.
- The Data Science Agent can now also use models served through Anthropic on Databricks
Uses models served through Anthropic when partner-powered AI features are enabled. Requires partner-powered AI features to be enabled.
- Mosaic AI Model Serving now supports OpenAI GPT 5 models (preview)
Supports GPT-5, GPT-5 mini, and GPT-5 nano models for batch inference with AI Functions.
- Anthropic Claude Sonnet 4.5 now available as a Databricks-hosted model
Accessible via Foundation Model APIs on a pay-per-token basis.
- Notebook improvements
Cell execution minimap appears in right margin, supports up to 100MB notebooks.
- Prompt caching is now supported for Claude models
Caches thinking messages, images, and tool use via cache_control parameter.
- Semantic metadata in metric views
Requires YAML 1.1 and Databricks Runtime 17.3 or higher.
- Databricks SQL version 2025.30 is now available in Preview
Adds LIKE operator support for UTF8 collations and ST_ExteriorRing function.
- Mosaic AI Agent Framework supports automatic authentication passthrough for Lakebase resources
Requires MLFlow 3.3.2 or above.
- Databricks Assistant Agent mode: Data Science Agent is in Beta
Orchestrates multi-step workflows from a single prompt, building entire notebooks for tasks like EDA and machine learning.
- Get this table by email. One Monday mail covering the week, filtered to the clouds and products you run. Weekly digest
- AI agents: Authorize on-behalf-of-user Public Preview
Lets agents act as the Databricks user who runs the query for added security.
- Admins can now manage a workspace's serverless base environments (Public Preview)
Defines custom environment specs for serverless notebooks, allows setting a default for new notebooks.
- Default warehouse setting is now available in Beta
Applied across SQL editor, AI/BI dashboards, and other tools, with optional user override.
- External MCP servers are in Beta
Enables agent access to external tools, requires connecting to Model Context Protocol servers.
- Databricks Assistant integrated with compute
Creates compute resources, pools, and policies, and answers questions about them. Available on some compute pages.
- Databricks SQL version 2025.25 is rolling out in Current
Rolls out from August 20th to August 28th, 2025.
- Databricks-hosted models for Assistant are now GA
Available on all cloud platforms at no extra cost.
- Databricks SQL version 2025.25 is now available in Preview
Adds recursive common table expressions and spatial SQL expressions.
- Fixed timeout handling for materialized views and streaming tables
New views and tables apply warehouse timeout, existing ones require CREATE OR REFRESH. Default timeout is 2 days.
- New SQL editor is generally available
Provides a unified authoring environment with features like real-time collaboration and enhanced Databricks Assistant integration.
- Token-based rate limits now available on AI Gateway
Configures rate limits on model serving endpoints, replacing legacy governance settings, and requires configuration updates.
- OpenAI GPT OSS models now support function calling
Supports function and tool calling on Databricks-hosted models.
- OpenAI GPT OSS models now supported on provisioned throughput
Supports GPT OSS 120B and 20B models on provisioned throughput.
- Schedule jobs for Serverless GPU compute workloads
Creates and schedules Serverless GPU jobs via notebook scheduling.
- Databricks Assistant: User and workspace instructions now available
Custom instructions can be added by users or admins, applying to all Assistant experiences.
- Use OpenAI GPT OSS models for batch inference and structured outputs
Optimized for AI Functions with batch inference and structured outputs. Supports ai_query().
- OpenAI GPT OSS models now available on Mosaic AI Model Serving
Supports GPT OSS 120B and 20B as hosted foundation models, accessible via pay-per-token APIs.
- Edit mode in Assistant does multi-cell code refactoring and more
Performs reasoning and editing across multiple cells without manual intervention.
- Preset date ranges for parameters in the SQL editor
Includes options like This week, Last 30 days, and Last year for timestamp and date parameters.
- Jobs & Pipelines list now includes Databricks SQL pipelines
Includes materialized views and streaming tables created with Databricks SQL.
- Inline execution history in SQL editor
Shows past results without re-executing queries, links to past query profiles.
- Databricks SQL version 2025.20 is now available in Current
Rolls out in stages to the Current channel, see 2025.20 features for details.
Headlines, dates and product areas are Databricks' own, and every item links to the note it came from. The one-line summaries are ours.