Databricks releases, every cloud
- Base environments are now supported on classic compute (Beta)
Manages Python dependencies using Databricks-provided or custom environments.
- Moonshot AI Kimi K2.7 is scheduled for retirement
Replaced by Kimi K3, retiring October 30, 2026.
- Unity Gateway support for compliance security profile standards
Enabled by default for workspaces with compliance security profile. Supports all Databricks compliance standards.
- The counter_diff SQL window function is now available
Calculates differences between consecutive cumulative counter values, available in Databricks Runtime 19 and above.
- Window measures on a numeric index column in metric views
Supports unitless offset and ranges on consecutive integer columns, useful for non-calendar periods like fiscal weeks.
- DeepSeek V4.1 Flash is now available on Unity Gateway
Supports text and image inputs, accessible via Foundation Model APIs.
- Run the change, do not just read it. Hands-on labs in your own Databricks workspace, graded when you submit. Browse labs
- Standard performance mode is now available for one-time runs in the Jobs API
Sets performance_target to STANDARD. Supports Apache Airflow's DatabricksSubmitRunOperator.
- Partner-powered AI features can no longer be disabled in the settings UI
Settings API still allows enable or disable until November 1, 2026.
- See which chunks support each ai_search answer (Beta)
Added generate_citations option to show supporting chunks.
- The partner-powered AI features setting will be removed Coming soon
Settings will remain as is, but changes will require contacting the account team after removal.
- time_bucket SQL function aligns timestamps to fixed-width time buckets
Replaces date_trunc for non-calendar units, requires origin and width parameters.
- OpenAI GPT-6 Astra is now available on Unity Gateway
Available for enterprise workloads via pay-per-token Foundation Model APIs.
- Environment version 6 is now available
Supports serverless and standard classic compute.
- AI Search supports struct and map columns and filter-only queries
Supports indexing and filtering of struct and map columns, and filter-only queries without vector retrieval.
- Git Folder Serverless is in Beta
Shares serverless compute and pyproject.toml environment across notebooks and files.
- Google Gemini 3.8 Flash now available as a Databricks-hosted model
Accessible via Foundation Model APIs, requires compliance with applicable terms.
- Anthropic Claude Fable 5.1 now available as a Databricks-hosted model
Accessible via Foundation Model APIs, requires compliance with applicable terms.
- Zhipu AI GLM 5.3 now available as a Databricks-hosted model
Supports up to 1,048,576 tokens and 65,536 output tokens. Accessed via Foundation Model APIs.
- Get this table by email. One Monday mail covering the week, filtered to the clouds and products you run. Weekly digest
- Unity Catalog Skills (Beta)
Publishes and governs agent skills, controlling access with Unity Catalog permissions.
- AI Functions REST API is now generally available
Includes ai_parse_document, ai_extract, and ai_classify calls.
- Zhipu AI GLM 5.3 Flash is now available on Unity Gateway
Access through the same governed API, with centralized permissions and no code changes required.
- ai_prep_search supports plain text and markdown input (Beta)
Accepts STRING content, no need for ai_parse_document preprocessing.
- Consumer access to query Unity Gateway services will become generally available Coming soon
Requires account admins to set budgets and rate limits beforehand to control model traffic.
- Default Python package repositories are GA for Lakeflow pipelines and classic compute
Supports private or authenticated registries as workspace defaults. Applies to UI and API created classic compute.
- Upcoming end of life for the Supervisor API Coming soon
Replaced by custom agents on Databricks Apps, ends September 30, 2026.
- ai_extract precision mode is now generally available
Supports multi-page documents and schemas with 50+ fields.
- Manage project environments with %uv commands
Requires environment version 5 or above. Supports serverless notebooks.
- Google Gemini 3.7 Flash is now available on Unity Gateway
Access through the same governed API as other models, with no application code changes required.
- Smart Routing: Cut AI costs by routing each task to the right model
Works with Claude Code and Codex via Unity Gateway CLI or Omnigent. Cuts task cost by over 30%.
- REPLACE USING flows for standalone streaming tables (Beta)
Replaces rows matching specified key columns, requires SEQUENCE BY column.
- Alert system tables are in Public Preview
Contains alerts and alert_evaluation_history tables for auditing and monitoring.
- Target an Azure capacity reservation group for classic compute
Requires VNet injection, guarantees compute capacity for constrained VM types.
- ai_search function (Beta)
Retrieves info from AI Search indexes, generating optimized queries and returning relevant documents.
- Change data feed on materialized views (Beta)
Requires Databricks Runtime 18 LTS or above. Enables replication to external destinations.
- JAR tasks on serverless compute are now generally available
Requires matching Scala, Java Development Kit, and Databricks Connect versions.
- Unity Gateway is now generally available
Controls AI service access and traffic, part of Unity Catalog, with some features still in Beta.
- Automatic cost attribution for materialized views and streaming tables in Databricks SQL
Inherits custom tags from enclosing SQL warehouse environment for billing.
- Serverless compute access control is now generally available
Governs access to notebooks, jobs, and pipelines through Default Interactive and Automated Compute objects. Existing workloads run unchanged.
- Web terminal on serverless GPU compute (AI Runtime) is in Public Preview
Runs shell commands, monitors GPU usage, and manages files on serverless GPU compute. Requires environment version 5 or above.
- Genie Code is now available on HIPAA compliance security profile workspaces with partner-powered AI features disabled
Genie Code uses Databricks-hosted models, with partner-powered AI features disabled.
Headlines, dates and product areas are Databricks' own, and every item links to the note it came from. The one-line summaries are ours.