Last week OpenAI launched a new version of its most advanced coding model, the GPT-5.2-Codex, which the company says was designed for complex software engineering tasks and also for defensive cybersecurity initiatives.
Based on the recently released GPT-5.2, the new release focuses on improving the management of large-scale projects, especially when code is distributed across multiple repositories and changes are introduced gradually over time.
Being able to maintain operational context more efficiently, without the need to use numerous tokens, this new model handles large refactorings, technology migrations, and structural changes in code in a more effective way, ensuring greater consistency across different stages while minimizing common errors, such as those that occur when an AI needs to operate on massive codebases in professional environments.
Compared to previous versions, Windows support has been expanded, something critical for companies that still maintain their infrastructure outside the Linux ecosystem.
Moreover, improvements were made to image processing features, making it simpler in terms of analyzing graphs, technical schematics, and user interface captures. Tool performance has also become more dependable, particularly during command-line operations and practical coding scenarios.
OpenAI’s GPT-5.2-Codex has shown remarkable results in standardized testing, scoring an impressive 56.4% on SWE-Bench Pro and an outstanding 64% on Terminal-Bench 2.0. This updated version is particularly exciting in security applications, with enhanced code analysis features that improve its capacity to detect security flaws.
While it hasn’t yet achieved a “High” cyber capability rating according to OpenAI’s Preparedness Framework, the company is proactively adding extra security measures to prepare for future iterations that might reach that level. The deployment is happening in phases, with current availability through Codex CLI, IDE plugins, cloud platforms, and code review applications for ChatGPT users, while broader API access for external developers is planned for later.
Additionally, OpenAI has introduced a limited-access initiative specifically for authenticated cybersecurity professionals and authorized entities with well-defined use cases.
