Python 3.14: Template Strings, Subinterpreters, and Deferred Annotations

Python 3.14, released in October 2025, brought several long-awaited features to the language. Here’s a look at the highlights.

Template String Literals (t-strings)

PEP 750 introduces t"..." literals that create template objects instead of plain strings. Unlike f-strings (which are evaluated immediately), t-strings produce reusable templates with deferred interpolation:

from string.templatelib import Template

name = "World"
t = t"Hello, {name}!"
# t is a Template object, not a string

str(t)  # "Hello, World!"

# Templates are safe by default — no injection risk
greet = t"Hello, {name}"
# Later, with different context:
name = "Python"
str(greet)  # "Hello, Python"

Deferred Annotation Evaluation

PEP 649 (with PEP 749 refinements) makes annotation evaluation lazy. Annotations are stored as functions and evaluated only when accessed via get_annotations():

def greet(name: str) -> str:
    return f"Hello, {name}"

# Annotations are not evaluated at definition time
# They're evaluated lazily when requested
import typing
typing.get_annotations(greet)  # {'name': str, 'return': str}

This means forward references work without from __future__ import annotations.

Multiple Interpreters in the Standard Library

PEP 734 adds the concurrent.interpreters module, letting you run Python code in isolated subinterpreters — each with its own GIL:

from concurrent.interpreters import Interpreter

interp = Interpreter()
interp.run("""
import math
result = math.factorial(100)
""")

This is especially useful for CPU-bound parallel work without the GIL limitations, since each subinterpreter has its own GIL.

Zstandard in the Standard Library

PEP 784 adds Zstandard compression support via compression.zstd:

import compression.zstd as zstd

data = b"Hello, World!" * 1000
compressed = zstd.compress(data)
decompressed = zstd.decompress(compressed)

Other Highlights

• Safe debugger interface (PEP 768): Debuggers can now attach to running processes without stopping them
• Tail-call interpreter: Up to 5% faster on supported compilers
• Improved error messages: More helpful tracebacks for common mistakes
• Free-threaded mode improvements: Performance penalty reduced to ~5-10%
• New annotationlib module for introspecting annotations programmatically

Python 3.14 is the current stable release, and 3.14.6 (June 2026) is the latest bugfix version.