Overview
DuckDB is an in-process embedded OLAP SQL database popular for local analytics, Parquet/CSV workloads, and lakehouse-style pipelines—often paired with motherduck for hosted scale-out.
Recent Developments
- 2026-09–10: v2.0 alpha (
v2.0-cyanoptera) with feature freeze; GA projected second half of October 2026 (2026-10-10-duckdb-2-0-performance-improvements) - Async remote I/O: Default read-ahead for S3 Parquet/CSV—~2–3× faster on large Parquet in vendor/MotherDuck benchmarks; up to ~20× on large CSV in DuckDB team tests
- Recursive CTE engine rewrite: Large graph traversals up to ~40× faster in cited LDBC-style workloads
- VARIANT type shredding: Faster semi-structured JSON queries vs. 1.5.x VARIANT; triggers;
quackextension protocol 1.0
Alpha builds are not production-ready; benchmarks are primarily DuckDB/MotherDuck reported.