What’s new in 3.0.3 (May 11, 2026)#

These are the changes in pandas 3.0.3. See Release notes for a full changelog including other versions of pandas.

Enhancements#

Starting with pandas 3.0.0, time zones are represented by default using the standard library’s zoneinfo module (or datetime.timezone for fixed offsets) instead of using pytz (release note).

The IO methods using pyarrow under the hood such as read_parquet(), read_feather() and read_orc() (or read_csv() when specifying the engine) were still returning timezone using pytz. Those have now been updated to consistently use default zoneinfo time zones as well (GH 65134).

Fixed a regression in pandas.bdate_range() where specifying end on a weekend with periods returned fewer dates than expected (GH 64834)
Fixed a regression in to_timedelta() ignoring the unit argument for round float values when mixed with non-round floats in a list (GH 65150)
Fixed a regression in Series.rank() with custom extension dtypes (GH 64976)
Fixed a regression in Timedelta.round(), Timedelta.floor(), and Timedelta.ceil() raising ZeroDivisionError for sub-second freq (GH 64828)
Fixed reading of Parquet files with timezone-aware timestamps or localizing of a timeseries with a pytz timezone when an older version of pytz was installed (GH 64978)
Fixed a regression in read_csv() with engine="c" where thousands="," could incorrectly parse non-numeric values like "1 ," as integers instead of strings (GH 64655)
Fixed a performance regression when working with timezone-aware timeseries data using a default zoneinfo time zone, i.e. slowdown when accessing components or localizing a naive timeseries (GH 64363)

Fixed ArrowDtype string arithmetic for addition between string and large_string typed arrays (GH 65220)
Fixed a bug in adding missing values (e.g. NA) to a pyarrow-backed string array or Series raising an error instead of propagating the missing value (GH 64968)
Fixed a bug in Series.rank() with period dtype and missing values, always sorting missing values at the top, regardless of the na_option value (GH 64976)
Fixed a bug in reading files with a chained URL scheme (e.g. tar://) with the latest version of fsspec (GH 65462).
Fixed a hang in Series.str.replace() for PyArrow-backed string data when replacing an empty pattern; behavior now matches Python’s str.replace (GH 64941).

A total of 10 people contributed patches to this release. People with a “+” by their names contributed a patch for the first time.