On October 21, 2026, the DuckDB Foundation and DuckDB Labs unveiled DuckDB 2.0.0 LTS, codenamed 'Cinnamon Teal' after the distinctive North American dabbling duck. Representing the culmination of nearly 20,000 commits since the 1.5 release series, version 2.0 marks DuckDB's first designated Long-Term Support (LTS) release, guaranteeing enterprise data platforms API stability, security backports, and bug fixes through October 2027.
While DuckDB originally gained industry prominence as an embedded, in-process analytical engine dubbed the 'SQLite of analytics,' version 2.0 significantly expands the system's operational topology. The release introduces native client-server networking powered by the Quack protocol. Using new SQL keywords such as CONNECT and DISCONNECT, applications can now attach to remote DuckDB instances over TCP/TLS, enabling collaborative multi-user analytics and pushing query optimizations down to external databases such as PostgreSQL and MySQL without third-party proxies.
Hardware saturation on modern cloud virtual machines served as a primary engineering priority for version 2.0. The maintainers implemented an asynchronous I/O engine engineered specifically for large cloud instances. On high-end configurations—such as AWS EC2 instances boasting 128 or more vCPUs—the new engine eliminates blocking thread bottlenecks when streaming Parquet files from Amazon S3 or Google Cloud Storage, delivering near-linear throughput scaling across massive analytical datasets.
The type system also received major architectural upgrades with the introduction of a first-class VARIANT data type. Designed to streamline semi-structured data manipulation, VARIANT permits dynamic, schema-flexible values within columnar tables, executing JSON-like nested filtering and type casting with negligible latency compared to full table flattening. Additionally, the release concludes the deprecation of legacy arrow-style lambda expressions in favor of clean, Python-style syntax.
To maximize storage density, DuckDB 2.0 transitions to storage format version 69. While DuckDB 2.0 remains capable of opening and querying database files written by older DuckDB versions, database files created or rewritten in format 69 are not backward-compatible with pre-2.0 releases. The DuckDB Foundation confirmed that precompiled binaries and language bindings across Python, R, Node.js, and Rust are immediately accessible on public package mirrors.