Free-Threaded Python and Subinterpreters

The Global Interpreter Lock (GIL) has been Python's defining constraint since its creation. Python 3.13 introduced the free-threaded build—a runtime without a process-wide GIL, instead using per-interpreter GILs to isolate state and eliminate contention. This fundamental shift unlocks true multi-threaded parallelism for CPU-bound workloads.

Subinterpreters extend this capability by allowing your Python process to run multiple independent interpreter instances, each with its own GIL and memory context. Combined with the interpreters module (PEP 554), subinterpreters let you spawn isolated Python environments that share minimal state, communicate via message-passing channels, and scale safely across all CPU cores without bottlenecks.

This series guides you from understanding the no-GIL architecture through building production-grade free-threaded applications. You'll learn when to migrate existing code, how to benchmark gains, design thread-safe semantics, and implement advanced patterns like fine-grained locking. Whether you're optimizing existing services or architecting new CPU-heavy systems, these articles equip you with the concepts and code to unlock Python's true parallel potential.

Articles in this series

• Python No-GIL: What It Means and Why It Matters
• How to Enable Free-Threaded Python in Your Project
• Understanding the Interpreters Module: Python's New API
• Subinterpreters vs Threads: When to Use Each
• Building Thread-Safe Code for Free-Threaded Python
• GIL-Free Data Sharing Between Subinterpreters
• Benchmarking Free-Threaded Python: Measuring Real Gains
• Per-Interpreter GIL: Isolating State and Avoiding Deadlocks
• Migrating Legacy Python Code to Free-Threaded
• Advanced Patterns: Fine-Grained Locking in Python