OpenAI introduces GPT‑5.2‑Codex

OpenAI has unveiled GPT‑5.2‑Codex, its latest agentic coding model designed to support complex, long-horizon software engineering and defensive cybersecurity tasks.

The company said GPT‑5.2‑Codex is further optimized for agentic coding in Codex, with improvements in context compaction, long-horizon work, large code changes such as refactors and migrations, better performance in Windows environments, and stronger cybersecurity capabilities.

According to the AI company, the model achieves state-of-the-art results on SWE-Bench Pro and Terminal-Bench 2.0, shows stronger vision performance for interpreting screenshots and technical diagrams, and is more reliable for multi-step agentic workflows in large repositories.

OpenAI said that, on 11 December, the React team published three security vulnerabilities after a researcher using GPT‑5.1‑Codex‑Max with Codex CLI and other agents reproduced and investigated a prior critical bug, and that this work led to responsible disclosure of the new issues.

The new model is available today across all Codex surfaces for paid ChatGPT users, the company said, with API access planned in the coming weeks and an invite-only trusted access pilot for vetted security professionals and organisations focused on defensive cybersecurity.

The Recap

• GPT‑5.2‑Codex is for professional coding and cybersecurity work.
• Available now for paid ChatGPT Codex surfaces users.
• API access is planned to roll out in the coming weeks.