OpenAI has released GPT-5.2-Codex, its most advanced agent coding model for real-world software engineering and defensive cybersecurity. It is built on GPT-5.2 and is designed to make long-form coding tasks more reliable and efficient.
The model is a special version built for Codex, focused on long-form coding tasks. It improves context handling, supports large code changes such as refactoring, and runs more reliably on large codebases and Windows systems.
OpenAI says GPT-5.2-Codex performs strongly in coding benchmarks and is aimed at practical use in debugging and complex projects. As the model’s capabilities grow, security controls will be in place, and access for cybersecurity teams will be limited and monitored.