Something not mentioned yet is that pandas is eager and copies memory quite liberally so even not too large datasets might blow up your RAM. Polars has an eager API but also a lazy one; I know they're also working on a query optimizer for lazy queries so you could theoretically handle even large data on a laptop for example.
Seconding this, the syntax feels SQL-like which I'm a fan of since I also sometimes have to write SQL queries. So they keep the mindset sharp and also are a clear paradigm to think in if I need to think about a polars dataframe.