Python 3.14: What’s New and Should You Upgrade?

What is new in Python 3.14: template strings, deferred annotations, supported free threading, multiple interpreters, zstd and what to check before upgrading.

Python 3.14 is a mature release whose headline changes are template strings, lazily evaluated annotations, subinterpreters in the standard library and an officially supported free-threaded build. It was released on 7 October 2025 and is supported until October 2030.

At a glance

Template strings (PEP 750)

A t-string uses a t prefix instead of an f. It keeps the static text and the interpolated values separate, so code can treat the values safely before joining them.

from string.templatelib import Template, Interpolation
import html

def render(template: Template) -> str:
    parts = []
    for part in template:
        if isinstance(part, Interpolation):
            parts.append(html.escape(str(part.value)))
        else:
            parts.append(part)
    return "".join(parts)

user_input = "<script>alert(1)</script>"
print(render(t"<p>Hello {user_input}</p>"))
# <p>Hello &lt;script&gt;alert(1)&lt;/script&gt;</p>
print(f"<p>Hello {user_input}</p>")
# <p>Hello <script>alert(1)</script></p>

The same approach applies to SQL, shell commands and log messages. A t-string is not a str, so passing one to code that expects a string needs a function like render in between.

Deferred evaluation of annotations (PEP 649 and 749)

Annotations on functions, classes and modules are stored and evaluated only when needed. You no longer need quotes around forward references, and the annotationlib module reads annotations in three formats:

from annotationlib import get_annotations, Format

class A:
    x: Undefined          # no quotes needed

get_annotations(A, format=Format.FORWARDREF)
# {'x': ForwardRef('Undefined', is_class=True, owner=<class '__main__.A'>)}
get_annotations(A, format=Format.STRING)
# {'x': 'Undefined'}
A.__annotations__
# NameError: name 'Undefined' is not defined

The What’s New page says that if you use from __future__ import annotations only to avoid quoting forward references, you may be able to remove it once you support only 3.14 and newer. Code that reads __annotations__ directly may need to move to annotationlib.get_annotations. See the type hints guide for annotating code.

Free-threaded Python is supported (PEP 779)

The free-threaded build moved from experimental to officially supported, which the release notes call phase II: supported but still optional. Whether it ever becomes the default is described as undecided. The implementation described in PEP 703 is finished, including the C API changes, and the performance penalty on single-threaded code is now roughly 5 to 10 percent, depending on platform and compiler (3.13 warned of a substantial hit). See free threading explained and what was in 3.13.

Multiple interpreters and Zstandard

The concurrent.interpreters module exposes a capability that was previously only available from the C API: several isolated interpreters in one process, with true multi-core parallelism.

from concurrent import interpreters

interp = interpreters.create()
interp.exec("print('hello from a subinterpreter')")
interp.close()

PEP 784 adds the compression.zstd module, and a compression package whose lzma, bz2, gzip and zlib modules re-export the existing ones:

from compression import zstd

data = b"python " * 1000
packed = zstd.compress(data)
print(len(data), "->", len(packed), zstd.decompress(packed) == data)
# 7000 -> 25 True

Compare the concurrency options in asyncio vs threading vs multiprocessing.

Smaller changes

• Brackets are optional in except (PEP 758). except TimeoutError, ConnectionRefusedError: is valid in 3.14 when there is no as clause. With as, Python 3.14 still reports multiple exception types must be parenthesized when using 'as'. On 3.13 the bracketless form is a SyntaxError.
• Syntax highlighting in the default interactive shell, and colour in several standard library command-line tools.
• A tail-calling interpreter, opt-in and needing Clang 19 or newer on x86-64 or AArch64. The release notes report a 3 to 5 percent geometric-mean speed-up on pyperformance.
• An experimental JIT in the official macOS and Windows binaries, enabled with PYTHON_JIT=1. The notes say it is not recommended for production use and that its effect ranges from 10 percent slower to 20 percent faster, depending on the workload.
• Garbage collector. Python 3.14.0 to 3.14.4 shipped an incremental collector. Because of reports of significant memory pressure in production, it was reverted to the generational collector from 3.13 in Python 3.14.5. If you read older articles about an incremental GC in 3.14, this is why they no longer apply.

What to check before upgrading from 3.13

• Removed APIs: the asyncio child watchers, ast.Num, ast.Str, ast.Bytes, ast.NameConstant and ast.Ellipsis, pkgutil.get_loader and pkgutil.find_loader, and urllib.request.URLopener and FancyURLopener.
• Multiprocessing: on Unix platforms other than macOS the default start method is now forkserver, not fork.
• Annotations: code that inspects __annotations__ directly.

The full list, with a test you can run, is in Python 3.13 end of life, which covers the move to 3.14.

3.14 or 3.15?

3.15 was released on 9 October 2026 with lazy imports and a faster experimental JIT; see Python 3.14 vs 3.15 and what is new in 3.15. On PyPI on 4 November 2026, NumPy, pandas, SciPy, Pillow, Matplotlib and scikit-learn publish wheels for both. If you are on 3.12 or older, 3.14 is a sound target; see Python 3.12 end of life.

How to try it

With uv: uv python install 3.14, then uv run --python 3.14 python. Test your project in a fresh virtual environment with python -W error -m pytest.