with recent GPT 5.2 releasedOpenAI updated its popular coding model Codex and other related models, bringing more agentic use cases into its scope.
GPT-5.2-codecs, called OpenAI in a blog post “The most advanced agentic coding model to date to engineer complex, real-world software,” it is optimized for long-horizon work with agents and will have strong cybersecurity capabilities.
This model is a branch of GPT-5.2, optimized for agentic building.
“The GPT‑5.2-Codec represents a step forward in how advanced AI can support real-world software engineering and specialized domains like cybersecurity – helping developers and defenders tackle complex, long-horizon work and strengthening the tools available for responsible security research,” the company said in its blog post.
Enterprises can use the new codec model “across all codec surfaces for paid ChatGPT users, and are working toward securely enabling access to GPT‑5.2-codecs for API users in the coming weeks.” The company is also piloting a program with invite-only trusted users to access “a more permissive model for vetted professionals and organizations” for defensive cybersecurity work to determine the balance between access and security.
Advances in Cyber Security with Models
OpenAI calls GPT-5.2-Codex its strongest cybersecurity model to date. Still, as its capabilities grow, the company said it needs to create a deployment approach that accounts for future growth and supports defensive cybersecurity.
“As our models continue to advance the intelligence frontier, we have seen that these improvements also translate into increased capability in specific domains such as cybersecurity,” the company said.
OpenAI said in its system card It tested the model on three benchmarks: capture-the-flag (CTF) evals, CVE-Bench, and Cyber Range.
GPT-5.2-Codex became the company’s strongest performing model in the CTF evaluation, which they attributed to compactness, or “the ability for the model to work coherently across multiple context windows.”
The model scored 87% in CVE-Bench and outperformed other models, with gpt-5.1-codecs-max coming in second. This enhancement will be helpful for tasks like running commands around vulnerability discovery and trying out tools “with an almost brute-force approach.”
In the long-form cyber range test, the models had a combined pass rate of 72.7%. GPT-5.1-Codex-Max scored 81.8%.
cyber security deployment project
OpenAI told some users of its GPT-5.1-Codex-Max, which was launched in NovemberExposed and subsequently reported a source code exposure vulnerability in React. According to OpenAI, Andrew McPherson, a security researcher at Privy, used GPT-5.1-Codex-Max to assess how well the model could support real-world vulnerability research. Instead the model revealed unexpected behavior.
With improvements in cybersecurity capabilities for the GPT-5.2-codecs and potentially subsequent models, OpenAI said it needs to balance the deployment of Frontier models with the tools needed for defensive cybersecurity. While the GPT-5.2-codec “does not reach the high level of cyber capability under our readiness framework,” the company plans to bring on selected users to test security capabilities. (OpenAI preparation outline To measure and track the potential harm caused to humans by AI)
“Security teams may face restrictions when attempting to simulate threat actors, analyze malware to support remediation, or stress test critical infrastructure. We are developing a Trusted Access pilot to remove that friction for qualified users and organizations and enable trusted defenders to use Frontier AI cyber capabilities to accelerate cyber defense.” OpenAI said.
agentic boundaries
GPT-5.2 already Received praise from users For its use in business functions and workflows. With the Codex version, some of those capabilities may be transferable, especially when enterprises plan to use the model to code their agents.
The company said the model improves long-horizon work through compaction, offering robust performance on extensive code changes. It also features better performance on Windows.
In benchmark testing, the GPT-5.2 codec performed best in terms of accuracy compared to its previous versions.
"With these improvements, Codex is more capable of working in large repositories over extended sessions with the full context. It can more reliably complete complex tasks like large refactors, code migrations, and feature builds – continuing to iterate without losing track, even when plans change or an attempt fails." OpenAI said.
since it launched Preview in MayCodex has helped bring acceptance of agentic and vibe coding into the enterprise AI builder space. With Windsurf, Cursor, Cloud Code, and Google’s many coding agents, the platform moved LLM from simple code completion to making it easier for users to create and launch asynchronous coding projects.
<a href