With the recent update of GPT 5.2, OpenAI has introduced significant improvements to its coding model Codex, specifically designed for complex software engineering tasks and enhanced cybersecurity capabilities. The new GPT-5.2-Codex model is optimized for long-horizon work with agents, offering a more robust solution for developers and defenders. OpenAI aims to make this model accessible to enterprises through various platforms, with plans to expand its availability further in the future.
Advances in cybersecurity with models:
OpenAI has labeled GPT-5.2-Codex as its most powerful cybersecurity model to date, showcasing its capabilities in various benchmarks such as Capture-the-Flag (CTF) evals, CVE-Bench, and Cyber Range. The model has excelled in CTF evals, scoring high in CVE-Bench, and delivering promising results in the Cyber Range test. These advancements highlight the model’s potential for cybersecurity tasks and defensive strategies.
Cybersecurity deployment project:
Following the launch of GPT-5.1-Codex-Max, OpenAI has been focusing on improving its cybersecurity capabilities, aiming to balance the deployment of frontier models with essential tools for defensive cybersecurity. The company plans to involve selected users in testing the security capabilities of GPT-5.2-Codex, ensuring a safe and efficient deployment process for cybersecurity tasks.
Agentic frontiers:
GPT-5.2-Codex has received positive feedback for its performance in business tasks and workflows, particularly in coding scenarios. The model’s enhanced capabilities, such as compaction and accuracy improvements, make it suitable for handling extensive code changes and complex tasks like refactors and code migrations. OpenAI envisions enterprises utilizing Codex for coding their agents, showcasing its potential in the AI builder space.
