🐍 The biggest pandas release in years just dropped — with breaking changes. Are you ready?
January/February 2026 was a busy month for the Python ecosystem. The headline: pandas 3.0, the first major release in years.
🔥 Key changes in pandas 3.0:
- String dtype by default: strings no longer infer to
objectbut tostr→ better performance and type safety - Copy-on-Write (CoW): all indexing operations return copies → breaks chained assignment
df[col][row] = value - New
pd.col()syntax: cleaner expressions for.assign()and similar methods - Anti-joins:
pd.merge(how="left_anti")now available - Python 3.11+ minimum required
⚠️ Changes that affect you:
# Before (pandas 2.x)
df['col']['row'] = value # WARNING then, ERROR now in 3.0
# Now (pandas 3.0)
df.loc['row', 'col'] = value # Correct📌 More news from the month:
- Python 3.15 alpha 5: JIT compiler improves 7–8% on AArch64 macOS
- PEP 822: proposes d-strings for cleaner multiline strings
- Anthropic invests $1.5M in PyPI security
- Polars 1.37: requires Python 3.10+
- PyTorch 2.10: deprecates TorchScript
💡 Explanation in a nutshell#
pandas 3.0 brings major performance improvements but also breaks existing code, especially “chained assignment” (modifying a DataFrame in two consecutive steps). This used to generate warnings; now it simply fails. The recommendation is to migrate to pandas 2.3 first, resolve all warnings, then upgrade to 3.0.
More information at the link 👇
Also published on LinkedIn.
