Snowflake October 2026 update: Postgres mirroring, the PERIOD type, Semantic Studio and new AI models

Snowflake ships something almost every day, which makes it hard to see what actually matters. This roundup covers the releases from mid-September to 16 October 2026, with the features most useful to analysts and analytics engineers first. Snowflake's big yearly event, EXPEDITION 2026, follows online on 3 to 5 November (Americas) and 4 to 6 November (other regions), so expect more agent news then.

Postgres data mirroring is GA

If your app runs on Snowflake Postgres, its tables can now be mirrored into Snowflake for analytics continuously. Inserts, updates, deletes and schema changes carry over, new tables in a mirrored schema are picked up automatically, and Snowflake says changes are visible within seconds. Under the hood, a Postgres extension (pg_lake) reads logical replication and writes Apache Iceberg tables, applied with serverless compute.

CALL SNOWFLAKE.POSTGRES.CREATE_MIRROR(
    mirror_name       => 'data_mirror',
    postgres_instance => 'my-pg',
    postgres_database => 'postgres',
    target_database   => 'MIRRORTOSNOWFLAKE',
    postgres_tables   => NULL,
    postgres_schemas  => ['public'],
    refresh_interval  => '1 minute',
    warehouse         => NULL
);

Each mirrored table has a $live view for the latest data and a $changes feed with seven days of history. It runs on AWS and Azure where Snowflake Postgres is available. For analysts, it means app data you can query with plain SQL without asking engineering for a nightly extract.

PERIOD: a real data type for date ranges

Contracts, subscriptions, employment, price validity, room bookings: so much business data is a range with a start and an end. Snowflake now has a PERIOD type for exactly that. Ranges are half-open, so the start is included and the end is not, which means back-to-back periods never overlap by accident.

CREATE OR REPLACE TABLE employment_windows (
  employee_id   INTEGER,
  active_period PERIOD(DATE)
);

INSERT INTO employment_windows VALUES
  (1, PERIOD(DATE) '[2020-01-15, 2023-06-30)'),
  (2, PERIOD_CONSTRUCT(DATE '2022-09-01', DATE '2024-01-01'));

-- Who was employed on 1 January 2023?
SELECT employee_id
FROM employment_windows
WHERE PERIOD_CONTAINS(active_period, DATE '2023-01-01');

Semantic Studio is GA

Semantic views describe your business in terms an AI can use: tables, relationships, facts, dimensions and metrics, plus synonyms and verified example questions. Semantic Studio, inside Snowsight Workspaces, is now the place to build them. You write a semantic view as YAML, preview a diff and deploy it, with optional Git branches and reviews. The Semantic View Autopilot wizard can start from SQL queries, a YAML spec, or an existing Tableau or Power BI file, and the built-in assistant can trace a bad agent answer back to the definition that caused it.

Also new: DEFINE SEMANTIC VIEW in DCM Projects is GA (30 September), so semantic views can live in the same deployment projects as your tables.

Cortex agents: code execution and an SDK

Cortex Agents plan a task, call tools and answer inside Snowflake's governance. Two upgrades this month:

• Code execution tool (GA, 5 October): agents can run Python in an isolated sandbox for calculations and data processing, alongside Cortex Analyst (SQL from questions) and Cortex Search (documents).
• Cortex Agent SDK (public preview, 13 October): build and run agents from your own code.

Around them: Horizon Context (preview) brings business context into Snowflake's governance catalogue, Agentic Marketplace Discovery is in public preview, and document generation in Snowflake CoWork reached GA on 14 October.

New AI models in Cortex

Response caching for Cortex AI functions also arrived at the end of September, which can cut cost when the same prompts run repeatedly, for example in a scheduled classification job. For a model-by-model comparison, see the best AI models in October 2026 and our explainer on Claude Haiku 5.5.

Smaller changes worth knowing

• Apache Iceberg v3: table upgrades and expression defaults are GA (16 October). Upgrade only when every engine that reads the table supports v3, because there is no downgrade. More in Iceberg v3 explained.
• Iceberg partition evolution is GA, and Snowpipe Streaming can write partitioned Iceberg tables.
• Dynamic tables keep refreshing incrementally after a failover, and incremental refresh now supports VALUES clauses and non-deterministic aggregates.
• Lineage is preserved through temporary tables and views, and now covers storage integrations, external tables and Cortex Search services.
• Zero-copy integration with Workday Data Cloud is GA, and stable egress IP addresses are GA on Azure.

Keep the Snowflake SQL cheat sheet handy, and if you are new, start with getting started with Snowflake.

Sources: Snowflake's release notes and feature updates (to 16 October 2026), the Snowflake blog post "Data Mirroring Is GA" (15 October 2026), Snowflake's PERIOD data type and Semantic Studio documentation, and Snowflake's EXPEDITION 2026 press release (5 October 2026).

Alessandro Ecclesie AgazziWritten by Alessandro Ecclesie Agazzi, analytics engineer in London. Updated 17 October 2026. About me