Python 3.15 and 3.10 EOL: Scheduled Interpreter Updates

This year's What's new guide was prepared by

Hugo van Kemenade, release manager for versions 3.14 and 3.15.

What Willison looked forward to most was explicit lazy imports from PEP 810: ```python lazy import json lazy from pathlib import Path ```

The interpreter registers the name but loads the module only when it is first accessed.

If the program does not reach `json` during that run, the module will not load.

The second notable new feature is a profiler that attaches to an already running process by PID.

That same day, Willison released datasette-ports 0.4: the plugin finds Datasette instances running on the machine and now displays each PID, making it easier to attach a profiler.

The library maintainer spent only a few hours supporting the new feature, which is typically what an ecosystem prepared for a release in advance looks like.

Startup time matters where a process starts frequently and runs briefly: CLI tools, cron jobs, serverless functions, and tools an AI agent calls dozens of times per session.

Heavy libraries such as pandas, cloud-provider SDKs, and LLM clients are imported on every run, even when a particular call does not need them.

The `python -X importtime` command shows how much this costs in your code.

Measure first, then promise a speedup: the result depends heavily on the project.

Before PEP 810, teams hid imports inside functions.

The technique worked, but it reduced readability, confused linters, and scattered dependencies throughout the file.

Now the dependency is declared at the beginning of the module, while loading time is spent only when it is used. The mechanism has a risk you should understand before migrating.

Lazy import defers import errors and module side effects until the first access.

If a library registers plugins or patches standard functions during import, the program's behavior will change.

A broken dependency used to crash the process at startup; now it will fail in production on a rare code path.

That is why `lazy` is introduced together with tests that exercise these branches.

Each Python version receives five years of fixes, with only security fixes in the final years. Starting in October 2026, 3.10 will not receive those either.

Library maintainers such as Willison are removing 3.10 from their tests and adopting 3.11 syntax. A company that remains on 3.10 will gradually become stuck on old dependency versions and have to address vulnerabilities on its own. There are also direct losses.

According to python.org, Python 3.11 is on average about 25% faster than 3.10 thanks to the Faster CPython project.

Version 3.11 was released in October 2022. A team still running 3.10 has gone four years without benefiting from a performance boost that requires no business-logic rewrite. Python 3.11 also introduced exception groups with `except*`, the built-in `tomllib` TOML parser, and error messages showing the exact location in a line.

The latter reduces incident investigation time.