What’s new in 3.0.1 (February 17, 2026)#

These are the changes in pandas 3.0.1. See Release notes for a full changelog including other versions of pandas.

Fixed regressions#

Added the pandas.api.extensions.ExtensionArray.item() method to fix regression with accessing the item of an array of strings, which now is an ExtensionArray subclass instead of a numpy ndarray (GH 63876)
Fixed regression in read_hdf() reading a file written by pandas <= 2.2 with a datetime64 column (GH 64006)
Fixed regression in to_timedelta() when converting a list with float and int (GH 64044)
Fixed regression in DataFrame.groupby() and Series.groupby() when grouping on categorical data with NA values, observed=False, and dropna=True (GH 52445)
Fixed regression in DataFrame.loc() when setting new row and new columns simultaneously filling existing columns with b'' instead of NaN (GH 58316)
Fixed regression in DataFrame.reindex() when using a string fill value (GH 63993)
Fixed regression in comparing a string column or index with a range object (GH 63429)
Fixed regression in comparison operation with a zero-dimensional subclass of np.ndarray (GH 63205)
Fixed regression in the DataFrame and DataFrame.from_records() constructor with a list of dicts with a missing value indicator as key (GH 63889)
Fixed regression when calling numpy.random’s permutation() on a (pyarrow-backed) string Series (GH 63935)
Fixed regression where methods such as Series.str.contains() and Series.str.replace() on pyarrow-backed strings rejected RE2 unicode escapes like \x{...} and \p{...} (GH 63901)

Fixed a bug in the DataFrame constructor when passed a Series or Index correctly handling Copy-on-Write (GH 63899)
Added additional typing aliases in pandas.api.typing.aliases (GH 64098)
Allow ExtensionArray to have dtypes involving numpy.void (GH 54810)
Fixed a bug in col() where unary operators (-, +, abs) were not supported (GH 63939)
Fixed bug in merge() where NaN values in pyarrow-backed string dtype join keys were incorrectly matched with non-NaN values on Windows with pyarrow 21 (GH 64060)
Fixed thread safety issues in DataFrame internals on the free-threaded build (GH 63685).
Prevent buffer overflow in Rolling.corr() and Rolling.cov() with variable windows when passing other with a longer index than the original window. This now raises ValueError (GH 62937)

A total of 57 people contributed patches to this release. People with a “+” by their names contributed a patch for the first time.