Tool Introduction
GPT‑5.2-Codex is the most advanced agent-based coding model to date, empowering complex software engineering practices. Built upon GPT‑5.2 and deeply optimized specifically for agent-driven coding, GPT‑5.2-Codex introduces several key enhancements: improved long-horizon task execution through context compaction, superior performance in large-scale code modifications (such as refactoring and migration), enhanced capabilities in Windows environments, and significantly strengthened cybersecurity defenses.
As our models continue advancing at the forefront of intelligence, we observe that these improvements translate into substantial capability leaps in specialized domains—particularly cybersecurity. Just last week, a security researcher leveraged GPT‑5.1-Codex-Max together with the Codex CLI to discover three vulnerabilities in React, which were responsibly disclosed (opens in a new window). These vulnerabilities could have led to denial-of-service attacks or source code leakage.
GPT‑5.2-Codex now features the strongest cybersecurity capabilities of any model to date. While these advances can significantly bolster cybersecurity at scale, they also introduce new dual-use risks that demand careful deployment. Although, according to our Preparedness Framework, GPT‑5.2-Codex has not yet reached the “High” threshold for cybersecurity capabilities, we are proactively designing deployment strategies with future capability growth in mind.
To balance innovation with responsibility, we are initially rolling out GPT‑5.2-Codex today via the Codex CLI, IDE extensions, cloud services, and code review tools—available immediately to all paid ChatGPT users. We are also actively working toward the secure release of an API to enable third-party access in the future. Additionally, we are developing a trusted-access mechanism specifically for cyber defense—a limited, invitation-only pilot program that grants vetted security professionals and organizations controlled, purpose-bound access to perform authorized defensive security operations under robust safeguards.
Pushing the Boundaries of Software Engineering
GPT‑5.2-Codex not only inherits GPT‑5.2’s strengths in professional domains but also integrates the cutting-edge agent-based coding and terminal operation capabilities of GPT‑5.1-Codex-Max. With enhanced long-context understanding, more reliable tool calling, higher factual accuracy, and native context compaction, GPT‑5.2-Codex becomes a more trustworthy partner for extended coding tasks while maintaining efficient token usage during reasoning.
GPT‑5.2-Codex demonstrates improved precision in interpreting shared screenshots, technical diagrams, data visualizations, and user interfaces during the coding process. Its agent-based coding performance is also notably more efficient and reliable in native Windows environments, further extending the frontier capabilities pioneered by GPT‑5.1-Codex-Max.
The model achieves outstanding results on benchmark evaluations such as SWE-Bench Pro and Terminal-Bench 2.0, which are designed to assess agent performance in real-world terminal environments across a variety of tasks.
Thanks to these advancements, GPT‑5.2-Codex significantly elevates performance in practical software engineering workflows—including codebase navigation, refactoring, and the creation and review of pull requests.