GPT-5.2-Codex (openai.com)

Today marked the announcement of GPT-5.2-Codex, the latest iteration of an agentic coding model, optimized for complex software engineering tasks. This update introduces significant enhancements such as long-horizon context compaction, improved performance on large code alterations, and stronger cybersecurity capabilities. The model excels in realistic terminal environments and native Windows operations, making it a powerful tool for developers. Notably, it also demonstrated advanced performance in identifying vulnerabilities, including a recently discovered exploit in React, showcasing its potential in cybersecurity domains. The release of GPT-5.2-Codex highlights the AI/ML community's growing focus on deploying advanced models safely while balancing accessibility with security. This version not only promises to assist developers in executing sophisticated coding tasks more efficiently but also emphasizes the importance of responsible AI deployment—particularly in cybersecurity, where higher capabilities increase the potential for misuse. The gradual rollout includes safeguards and an invite-only pilot for verified security professionals, reflecting a commitment to ethical applications of AI in real-world software engineering and defense against cyber threats.