OpenAI Launches GPT-5.2-Codex for Agentic Engineering and Cyber Defense

OpenAI has officially launched GPT-5.2-Codex, its most sophisticated agentic coding model designed for high-level software engineering and defensive cybersecurity. Released on December 18, 2025, this specialized version of the GPT-5.2 architecture targets professional developers managing complex repositories. The model is currently available to paid ChatGPT users, with an API rollout planned for the following weeks. This iteration introduces critical upgrades such as context compaction, which enables the model to handle long-horizon tasks without losing track of previous logic. It shows state-of-the-art results on the SWE-Bench Pro benchmark by generating accurate patches for large-scale codebases. Developers can also utilize its improved vision performance to turn design mocks into functional prototypes directly within the interface. A central feature of this release is the model's advanced proficiency in cybersecurity. GPT-5.2-Codex demonstrated a significant performance spike in professional Capture-the-Flag challenges, reflecting a growing ability to identify and mitigate vulnerabilities. OpenAI highlighted a recent case where a researcher used its predecessor to find a critical React vulnerability, underscoring the model's role as a force multiplier for security teams.

For the first time, OpenAI has significantly optimized model performance within native Windows environments. This expansion allows the agent to execute complex terminal commands, set up servers, and perform refactors across diverse operating systems. By improving tool-calling reliability and factuality, Codex acts more like an autonomous partner than a simple text assistant during extended coding sessions.

To manage the risks associated with these powerful tools, OpenAI is launching an invite-only Trusted Access pilot. This program provides vetted security professionals with access to more permissive versions of the model for ethical red-teaming and malware analysis. The goal is to ensure that defensive capabilities remain ahead of potential misuse as AI reaches new intelligence frontiers.