Launching today
Polars 2.0 is the biggest upgrade yet to the fast, open-source DataFrame library. Lazy queries now use its streaming engine by default, with initial out-of-core support that spills large workloads to disk instead of running out of memory. SQL is now a first-class citizen, backed by major optimizer and engine improvements, plus a new Map type and stricter APIs designed for faster feedback from both humans and AI coding agents.
Polars 2.0 is out today.
They originally said this would be a pretty boring major release, but quite a lot ended up making it in.
The streaming engine is now the default, there's initial spill-to-disk support for workloads that don't fit in memory, SQL is becoming first-class, and there are a bunch of optimizer and performance improvements under the hood.
They also published benchmarks against DuckDB and DataFusion, with Polars coming out fastest on almost every test they ran.
Pretty big milestone for a project that's now passed 775M downloads.