OpenAI has rolled out the latest version of its GPT Codex software coding AI model with improvements for complex programming work as well as enhancements that demonstrate how advanced AI can support specialized cybersecurity functions.
In the recent past, we analyzed GPT Codex enhancements that were designed to support long, complex coding sessions involving a range of tasks. With the latest release, GPT-5.2-Codex, OpenAI is driving further into complex coding.
These advances are important given the increasing reliance of software developers on AI to accelerate their work to deliver code faster and automate ever more complex tasks so they can place more focus on the most complex problems and coding functions, all while working in a secure software development environment.
Coding Enhancements
GPT-5.2-Codex delivers improvements on long-horizon work through context compaction (which shortens tool outputs to fit within a model’s token limit), stronger performance on large code changes like migrations and refactors of existing code for higher performance, as well as improved performance in Windows environments.
More specifically, GPT‑5.2-Codex has been enhanced in terms of long-context understanding, reliable tool calling, its level of factuality, and native compaction. These enhancements make it more dependable for engineers relying on GPT‑5.2-Codex to operate as a dependable partner during long-running coding tasks, while efficiently using tokens in the reasoning process.
GPT‑5.2-Codex delivers high performance on two key benchmarks, namely SWE-Bench Pro and Terminal-Bench 2.0, that test agentic performance on realistic coding tasks in terminal, or text-based, environments. It is also more effective and reliable at agentic coding in native Windows environments, building on capabilities introduced in GPT‑5.1-Codex-Max.
GPT‑5.2-Codex is more robust when it comes to working in large repositories over extended sessions while maintaining full context. It can more reliably complete complex tasks including complex refactors, code migrations, and feature builds — continuing to iterate without losing track, even when plans change or attempts fail.
Enhanced computer vision performance enables GPT‑5.2-Codex to more accurately interpret screenshots, technical diagrams, and charts as they are shared during coding sessions. As a result of these enhancements, Codex can take design mocks and quickly translate them into functional prototypes.
Cybersecurity Upgrades
OpenAI said that GPT‑5.2-Codex has stronger cybersecurity capabilities than any model released to date; the advances strengthen cybersecurity but they raise new dual-use risks that require careful deployment, the company noted.
GPT‑5.2-Codex does not reach a ‘High’ level of cyber capability under its Preparedness Framework for tracking risks; OpenAI said it’s got a sharp focus on expanding capabilities and is preparing for future models that do cross that threshold. OpenAI has added and documented additional safeguards in Codex.
Because security teams can run into restrictions when attempting to emulate threat actors, analyze malware to support remediation, or stress test critical infrastructure, OpenAI is developing a trusted access pilot to remove friction and enable qualified organizations and trusted defenders to use advanced AI-powered cyber capabilities to accelerate their cyber defense work.
OpenAI’s pilot program will initially be invite-only for vetted security professionals with a track record of vulnerability disclosure and organizations with clear, professional cybersecurity use cases. Qualifying participants will get access to OpenAI’s most capable models for defensive use cases.
GPT‑5.2-Codex is available today for paid ChatGPT users.
AI Agent & Copilot Summit is an AI-first event to define opportunities, impact, and outcomes with Microsoft Copilot and agents. Building on its 2025 success, the 2026 event takes place March 17-19 in San Diego. Get more details.
Tom Smith
Analyst
Cloud Wars, Agent and Copilot
Related Posts
