OpenAI just dropped GPT-5.2-Codex, their most advanced coding model yet, purpose-built for professional software engineering and defensive cybersecurity. It’s a tuned version of GPT-5.2 that excels at long-running, complex tasks like large refactors, code migrations, and feature builds, all while keeping full context intact through a new ‘context compaction’ trick. Performance in Windows environments also got a serious upgrade. Beyond coding, the model packs significantly stronger cybersecurity capabilities — so much so that a researcher recently used a previous version to find and responsibly disclose a React vulnerability that could have exposed source code. While GPT-5.2-Codex doesn’t hit ‘High’ risk under OpenAI’s Preparedness Framework, they’re already planning careful deployment for future growth. It’s rolling out today to paid ChatGPT users in Codex, with API access coming in weeks. An invite-only pilot for vetted defensive cybersecurity pros will also test more permissive versions. Why it matters: This release signals how quickly frontier AI models are becoming dual-use tools — equally powerful for building secure systems and for probing weak spots, forcing companies to balance access with safety like never before.