OpenAI presents by unveiling the GPT-5.2-Codex which is the most recent model in OpenAI’s Codex family (not the same as their Codex CLI or Codex Cloud coding agent tools).
GPT-5.2-Codex is a version of GPT-5.2 that has been further optimised for agentic coding in Codex. This includes enhanced cybersecurity capabilities, better performance in Windows environments, enhanced long-horizon work through context compaction, and improved performance on large code changes like migrations and refactors.
Similar to several earlier Codex models, this one is now accessible through their Codex coding agents and will be added to the API “in the coming weeks.” For “more permissive models,” there is a new invite-only preview procedure for qualified cybersecurity experts, in contrast to earlier models.
The capacity of GPT 5.2 to handle multi-hour agentic coding tasks has recently really amazed me. On the Terminal-Bench 2.0 benchmark, 5.2 Codex receives a score of 64%, whereas GPT-5.2 receives a score of 62.2%. It’s unclear how significant a 1.8% improvement would actually be.
Rather than hacking together API access again with earlier attempts from a users, I just asked Codex CLI to “generate an SVG of a pelican riding a bicycle” using the new model with medium effort. Here’s the transcript from a users attempt with the new Codex CLI timeline viewer and the pelican it came up with, below.
According to OpenAI, enhancing security is essential to AI-driven software engineering since dependable software is required for contemporary enterprise infrastructures. When it comes to identifying and resolving complicated software vulnerabilities, developers and security teams need all the assistance they can get. They also need to make sure that the AI coding tools being utilised don’t produce more vulnerabilities.
Earlier past month, security researcher Andrew MacPherson utilised GPT-5.1-Codex-Max to investigate the CVE-2025-55182 vulnerability in React, highlighting Codex’s software repair capabilities. With an explanation, he described in a blog post how the model mitigated the problem by combining fuzz testing, exploit analysis, and iterative assessments. In the process, it also revealed and fixed vulnerabilities that had not been discovered before.
OpenAI said GPT-5.2-Codex brings meaningful benefits for enterprises by automating complex, repetitive engineering tasks, supporting advanced application features, and strengthening cybersecurity operations to improve efficiency and reduce human error.
All paying ChatGPT customers can now access GPT-5.2-Codex, according to the business. In the upcoming week, it intends to expand access to application programming interface users and initiate a trial program for invite-only trusted access for qualified security experts with an emphasis on defensive cybersecurity.
We use cookies to ensure that we give you the best experience on our website. If you continue to use this site we will assume that you are happy with it.
