OpenAI releases GPT‑5‑Codex: Improved AI coding agent for software engineers

OpenAI has launched GPT‑5-Codex, an improved AI coding agent that is specifically targeted for use by software engineers, and it has been trained to conduct code reviews and find flaws.

It comes after the company's prior launch of Codex CLI and the cloud-based Codex. GPT-5-Codex can now run in the terminal, IDE software such as VS Code, in GitHub, on the cloud, and in the ChatGPT mobile app. ChatGPT Plus, Pro, Business, Edu, and Enterprise subscribers have access to GPT-5-Codex.

The OpenAI coding agent has been improved in assisting with many programming tasks, including creating websites, providing useful and correct code comments, and reviewing code for flaws. At the same time, its performance has been improved, including up to a 90% reduction in new task and follow-up completion times for certain cloud-based tasks.

GPT-5-Codex completes all of its work in a sandboxed environment. Access to the network and other tools is disabled by default, but can be activated by developers to suit their needs. When given web access, the AI coding agent can surf the web, build and iterate projects, and conduct code reviews on its own.

Because of its ability to conduct tasks on its own and its inherent vulnerability to hacks, the AI can be manipulated to conduct nefarious tasks that override its commands. As a result, OpenAI has rated GPT-5-Codex as High capability in the Biological and Chemical domains, activating further safeguards within the company to detect and counter its use to harm people.

The word codex is Latin for book, so those interested in seeing what ancient manuscripts looked like can peruse The Dresden Codex, sold here on Amazon.