OpenAI Released the New Coding Model GPT-5.2-Codex

GPT-5.2-Codex Architecture and Performance

OpenAI announced the GPT-5.2-Codex model, based on the GPT-5.2 infrastructure and optimized for complex software engineering processes. The new model includes major improvements especially in context compression, code refactoring, and performance in Windows environments.

In tests, the model outperformed previous versions, reaching an accuracy rate of 56.4 percent in the SWE-Bench Pro test and 64.0 percent in the Terminal-Bench 2.0 test. While the model’s visual capabilities can analyze technical diagrams and UI interfaces, it also has the capacity to turn design drafts into functional prototypes.

Technical Capabilities and Cybersecurity

GPT-5.2-Codex prevents performance loss by preserving context in long-term and multi-step tasks. The model’s notable technical capabilities are as follows:

• Restructuring: Optimizing the codebase to reduce memory usage or improve response times.
• Large-Scale Management: Managing code migration, feature creation, and large-scale editing operations.
• Advanced Security: AI-powered bug detection and the ability to identify vulnerabilities.

Although the model is not yet in the “High” cyber capability class under OpenAI’s Preparedness Framework, it has proven its effectiveness in practical scenarios such as detecting vulnerabilities in React. GPT-5.2-Codex is currently available to paid ChatGPT users, and API access is planned to open in the coming weeks.