The company states that the new model performs more effectively in handling cybersecurity-related tasks.
OpenAI has introduced the next model in the “Codex” series for programming, based on GPT-5.2. This model is optimized for long-running sessions. For example, it delivers better results in processes such as code refactoring and maintaining long-context memory.
According to OpenAI, GPT-5.2-Codex has the most advanced cybersecurity capabilities among the company’s existing models to date. However, the company warns that these capabilities could also be misused to identify system vulnerabilities and carry out cyberattacks (hacking).
In benchmarks such as “SWE-Bench Pro” and “Terminal-Bench 2.0,” where AI agents solve practical tasks in repositories and terminal environments, the new model improved performance by 1–2% compared to its predecessor, GPT-5.1-Codex.
GPT-5.2-Codex will be added to the built-in “Codex” AI agent within ChatGPT, which is available to all paid users. The model will also be integrated into the open-source “Codex CLI” agent designed for terminal use. In the coming weeks, it will also be released via the API.
