OpenAI’s New Codex Model

On December 18, OpenAI launched the next-generation Codex model, GPT-5.2-Codex, just a week after the release of the GPT-5.2 series. This model is touted as the most advanced coding model, focusing on professional software engineering and defensive cybersecurity, further solidifying OpenAI’s competitive advantage over Google’s Gemini.

OpenAI claims that GPT-5.2-Codex has made breakthroughs in coding performance, cybersecurity capabilities, and long-duration task handling. In the SWE-Bench Pro test, it achieved an accuracy of 56.4%, while in the Terminal-Bench 2.0 test, it reached 64.0%, setting new records for both benchmarks. The model was made available to paid ChatGPT users on the day of its release, with API access being rolled out progressively.

Enhanced Cybersecurity Features

OpenAI emphasized the significant improvements in cybersecurity with GPT-5.2-Codex. CEO Sam Altman noted that earlier this month, a security researcher used the previous model, GPT-5.1-Codex, to responsibly disclose a vulnerability in React that could expose source code. Although the new model has not yet reached a