• Dive statuses: Dives now carry a status (draft, ready, endorsed, or archived) to distinguish work-in-progress from vetted, endorsed Dives. Dive status shows up in the Dives list and the MCP/SQL tools for filtering and sorting.

• Inline Docs are now available in the Query Editor: Notebook cells in the MotherDuck query editor provide function information on hover, showing function signatures, parameter types, return types, and descriptions without leaving the notebook. Inline Docs can be toggled on and off by going to the Preferences page in Settings.

DuckDB 1.4 delivers performance gains with improvements like a rewritten sorting engine, more efficient small writes, and new SQL syntax including the MERGE statement. Learn more in the DuckDB 1.4.0 and 1.4.1 changelogs.

Performance improvements

• Sorting is 2x+ faster: Complete rewrite of sorting uses less memory and scales better across threads for ORDER BY, window functions, and list sorting
• More efficient small writes: Appending small numbers of rows now writes far fewer bytes
• 5x faster checkpointing: Reuses table metadata when tables aren't altered during checkpoint
• Parallel connection creation: Connections from instance cache can be created in parallel
• Faster scalar functions on dictionary data: Functions on dictionary-compressed data only run once per unique value

SQL syntax updates

• MERGE INTO statement: Standard SQL upserts without requiring primary keys or indexes
• FILL() window function: Interpolate missing values in ordered data
• Python-style macro arguments: Macros accept positional or named arguments for any parameter
• STRUCT to MAP cast: Direct casting between struct and map types

Parquet improvements

• VARIANT type reading: Read Parquet VARIANT types for faster semi-structured data processing
• Native geometry type writes: Write native Parquet geometry types
• Auto-globbing for directories: Automatically treats paths as directories and retries with glob patterns when no file is found

Learn more in the official DuckDB Labs announcements for 1.4.0 and 1.4.1.

While you can continue using your current version of DuckDB with MotherDuck, we encourage you to upgrade your DuckDB clients to 1.4.1 as soon as you can to take advantage of the fixes and performance improvements.

[Preview] DuckLake 0.3

As we announced earlier this year, MotherDuck now supports DuckLake, an integrated data lake and catalog format. DuckLake 0.3 makes working with DuckLake more robust, including CHECKPOINT for easy maintenance, new paths for Iceberg interoperability, spatial geometry types, and MERGE INTO support.

Learn more about using DuckLake databases in MotherDuck in the documentation, and the recent improvements in the DuckDB Labs announcement for DuckLake 0.3.

• New UX for FixIt: FixIt now includes keybindings for the toggles to accept and reject suggestions and turn automatic suggestions on and off.
  • Cmd/Ctrl + Enter: Accept suggestion and run query
  • Cmd/Ctrl + Shift + Backspace: Reject suggestion

• MotherDuck currently provides two options for creating and integrating with DuckLake databases:
  • Fully managed: MotherDuck manages both data storage and metadata
  • Bring your own bucket (BYOB): Connect your S3-compatible object storage with options for:
    • MotherDuck compute + MotherDuck catalog
    • Bring-your-own-compute (BYOC) + MotherDuck catalog

Learn more in the documentation and announcement blog.

DuckDB 1.3.0 improves performance in real-world scenarios for faster queries, new SQL syntax, and smarter Parquet file handling. Learn more in the changelog here.

Parquet improvements

• New TRY() expression for safer queries: More graceful handling for bad data by returning NULL instead of an error on problematic rows
• Pushdown of arbitrary expressions into scans: Reductions in unnecessary data processing to deliver up to 30x faster queries
• Pushdown of inequality conditions into joins: Major speedups for incremental dbt models and join-heavy queries

SQL syntax updates

• Python-style lambda syntax: You can now use lambda x: x + 1 instead of x -> x + 1; the old syntax is deprecated, but still supported.
• cast_to_type() function: Dynamically cast values to match column types - useful in generic expressions and CASE statements when writing macros.
• Recursive JSON access: New json_each() and json_tree() functions make it easier to traverse nested JSON structures.
• Struct field updates: Individual fields in structs can now be modified using ALTER; all fields are rewritten even if only one is updated.
• Prepared statements metadata: The duckdb_prepared_statements() function returns all prepared statements in the session.
• More flexible type definitions: Support has been added for CREATE OR REPLACE TYPE, CREATE TYPE IF NOT EXISTS, and CREATE TEMPORARY TYPE.
• Preserved order for OR filters: Execution now preserves the order of clauses in WHERE conditions using OR.
• Function alias visibility: duckdb_functions() now returns aliases in addition to the function outputs.

Parquet improvements

• Late materialization: Queries are 3–10x faster with LIMIT due to deferred column loading
• ~15% average speedup on reads: New scan and filter efficiency improvements
• 30%+ faster write throughput: Improved multithreaded export performance
• Better compression for large strings: Large string values are now dictionary-compressed
• Smarter rowgroup combining: Files are more efficient due to merging small rowgroups at write time

Learn more in the official DuckDB Labs 1.3.0 announcement.

While you can continue using your current version of DuckDB, we encourage you to upgrade your DuckDB clients to 1.3.0 as soon as you can to take advantage of the fixes and performance improvements.

Additional updates from this release are outlined below -

• Query results now display in a redesigned table that delivers enhanced performance when viewing and exploring data - column headers now include type information for better context. Additional table functionality, including sorting and filtering of results, is coming in future releases.

For earlier updates, see the release notes archive.