OpenAI Unleashes GPT-5.2 Codex for Advanced Coding & Cyber Defense

OpenAI has unveiled GPT-5.2-Codex, positioning it as their most advanced agentic coding model designed for complex, real-world software engineering. This iteration significantly boosts cybersecurity capabilities and is optimized for long-horizon tasks undertaken by AI agents. The model aims to provide developers and defenders with superior tools for tackling intricate challenges and conducting responsible security research.

Initial testing shows GPT-5.2-Codex performing strongly on benchmarks like Capture-the-Flag evals and CVE-Bench, outperforming previous versions in identifying vulnerabilities. While its performance in Cyber Range tests was slightly lower than GPT-5.1-Codex-Max, its overall advancements in code understanding and long-session coherence are notable.

Enterprises can expect phased access, beginning with paid ChatGPT users, followed by API access in the coming weeks. OpenAI is also piloting a program for vetted professionals to explore more permissive models for defensive cybersecurity work, balancing accessibility with crucial safety considerations.