Two weeks after Snowflake Expedition, the release notes from 5 to 16 November 2026 filled in many of the details. Some features are generally available (GA), others in preview. Our October update covers the month before.
AI functions: build, test, then pick the cheapest model that works
Snowflake's AI functions (AI_CLASSIFY, AI_EXTRACT, AI_COMPLETE and friends) let you run models from SQL. The hard part was always choosing the prompt and the model. Two releases on 16 November automate that.
- Cortex AI Function Studio (GA)
- Describe the task in natural language in Cortex Code CLI, or use the no-code flow in Snowsight AI Studio. It builds the function, benchmarks it with metrics you choose and tunes prompts and model choice. Works on text, documents, images, audio and video.
- AI Function Optimization (GA)
- A new experiment type, AI_FUNCTION_OPTIMIZATION, runs several prompt and model candidates on the same dataset and metric and reports quality and estimated cost. Save the winner as a governed function with CREATE AI FUNCTION, or use it in an AI_COMPLETE call.
- Pegasus 1.5 in AI_COMPLETE (GA)
- Twelve Labs' video model is now available through AI_COMPLETE.
- Page filtering for documents
- AI_EXTRACT and AI_PARSE_DOCUMENT can process only chosen pages with FL_FILTER_PAGES, which saves tokens on long files.
Semantic views get faster and handle many-to-many
Semantic views store your business definitions (metrics, dimensions, relationships) so that people and AI agents get the same answer. Two updates:
- Materializations are GA (11 November). They pre-aggregate the dimensions and metrics people query most. A bundled Cortex Code skill,
suggest-sv-materializations, looks at slow queries or a dashboard and recommends what to materialize, and two system functions tell you whether a query can use one. Semi-additive metrics, such as account balances, can now be materialized too. - Many-to-many relationships are in preview (13 November), through a bridge table or directly. Snowflake removes duplicates automatically, so totals are not double-counted.
-- A direct many-to-many relationship: one promotion code appears on many orders,
-- and the promotions table has no primary key
RELATIONSHIPS (
orders_to_promotions AS orders(promo_code) REFERENCES promotions(promo_code)
)One limit: queries that cross a many-to-many relationship cannot use level-of-detail metrics or multi-path metrics with the USING clause.
Security and cost controls
-- Temporary role grant (preview): access ends by itself after 24 hours
GRANT ROLE oncall TO USER alice FOR INTERVAL '24 HOURS';
-- SHOW GRANTS and ACCOUNT_USAGE.GRANTS_TO_USERS now include an expires_on column
SHOW GRANTS TO USER alice;Generally available
Use now- Per-user quota overrides: give one user a higher or lower credit limit, with an optional end date (11 Nov)
- Choose per-user quota users by role (11 Nov)
- Programmatic notifications for Trust Center findings (11 Nov)
- Openflow in AWS government regions (11 Nov)
Useful for keeping AI spending per person under control.
Preview
Try first- Temporary role grants with FOR INTERVAL or UNTIL (10 Nov)
- INTEGRATIONS view in ACCOUNT_USAGE (10 Nov)
- Organization users and groups in Organization Command Center (5 Nov)
- PLAN privilege for DCM Projects (12 Nov)
Temporary grants alone remove a whole class of forgotten access.
Tableau to Streamlit with Cortex Code
A new Cortex Code skill, /tableau-to-streamlit-in-snowflake (preview, 12 November), turns a Tableau workbook into a multi-page Streamlit in Snowflake app, one page per dashboard. Worksheets become charts and tables, calculations become Snowflake SQL, and filters and parameters become Streamlit widgets.
- 1
Review on fake data first
The app starts on synthetic data. Check the layout and the calculations before anything else.
- 2
Approve the switch to real tables
Cortex Code only connects your real tables after you approve the mapping.
- 3
Watch who sees the preview
Roles given access to the synthetic preview keep it after the switch, so share it only with roles allowed to see the real data.
- 4
Compare numbers
Check a few totals against the original Tableau report before you retire it.
Should you migrate at all? Our Power BI vs Tableau comparison covers the wider choice.
Also new
- Models: Claude 5.5 and GPT-6 models became generally available in Cortex on 6 November.
- Cortex Code: skill recommendations and easier skill sharing (9 November), and desktop app versions 1.25.0 and 1.25.4.
- Snowflake CoWork: geospatial maps in preview (10 November).
- Data Clean Rooms: analysis results can now be saved to a table in the clean room (12 November).
- AI_TRANSCRIBE: better transcription accuracy (13 November).
Check your understanding
1/3What does AI Function Optimization help you find?
It compares candidates on the same dataset and metric, with quality and estimated cost.
2/3How do you give someone a role for one day only (preview)?
The grant ends by itself; SHOW GRANTS shows an expires_on column.
3/3What do semantic view materializations do?
A Cortex Code skill can even suggest which ones to create.
You got 0 out of 3.
More: the Snowflake cheat sheet, Snowflake secondary indexes explained and BigQuery vs Snowflake.
Sources: Snowflake server release notes and feature updates (5 to 16 November 2026), including the release notes for semantic view materializations, many-to-many relationships, temporary role grants, per-user quota overrides, Cortex AI Function Studio, AI Function Optimization and the Tableau to Streamlit skill. Preview features may change before general availability.
