Release of Polars 2.0

Polars 2.0 has been released, bringing a host of new features and improvements. One of the most notable enhancements is the treatment of SQL as a first-class citizen, with significant advancements in Polars SQL coverage. The release also includes numerous optimizations to the engine and optimizer, such as query reordering, common-subplan elimination, and dynamic predicates/bloom filters.

To evaluate performance, Polars SQL was tested against TPC-H and TPC-DS data against DuckDB 1.5.6, DuckDB 2.0 alpha, and DataFusion (54.0.0) on c7a.4xlarge and c7a.metal machines. All engines completed all queries; however, Polars and the two DuckDB versions emerged as the fastest. Polars also scales well with increased cores, with a 3.8x improvement on TPC-H and 2.2x improvement on TPC-DS compared to DuckDB 1.5.6 and the DuckDB 2.0 alpha.

Some key features of Polars 2.0 include the default streaming engine for collect, enhanced out-of-core capabilities, and Arrow MapType support. The release also emphasizes stricter error handling and faster feedback for AI-driven development. Although DataFusion experienced timeouts on certain queries, Polars remains the fastest engine in most benchmarks. The complete benchmark details and a repository for replication are provided in the appendix.