OpenAI has launched the latest AI model, GPT-5.2-Codex, designed as an autonomous software agent for tackling complex tasks and identifying vulnerabilities. An exclusive access program allows verified experts to utilize a version with relaxed security filters. The model employs advanced context compression to efficiently manage long conversations and extensive code analyses, building on its predecessor, GPT-5.1-Codex-Max. Image processing has been enhanced for better interpretation of technical diagrams and user interfaces. Despite modest advancements in standardized benchmarks, such as SWE-Bench Pro and Terminal-Bench 2.0, the new model features a solution rate of 56.4% and a slight accuracy increase to 64%. With a strong focus on cybersecurity, GPT-5.2-Codex presents both defensive and offensive capabilities, necessitating a trusted access program for certified security experts. The model is now available to paying ChatGPT users, with integration via command line and cloud interfaces.
Share
Read more