OpenAI Launches Codex: Your New AI Coding Assistant

OpenAI has introduced Codex, a research preview of a specialized AI agent designed to assist with programming. Built on the advanced codex-1 model, a refined version of OpenAI's o3 reasoning model, Codex can generate, analyze, and test code across various tasks. The agent lives inside the ChatGPT web app, accessible through a new sidebar. Users can type a request and choose either "code" to generate code or "ask" to get technical advice and answers.

Codex runs each task in a container loaded with the user's codebase. This setup mirrors the user's real development environment, helping ensure accuracy and consistency. Developers can even include an AGENTS.md file in their repository—similar to a README, but specifically meant to guide AI agents.

Code You Can Understand

Codex writes code and shows its reasoning and intermediate results step-by-step. Depending on complexity, these processes can take anywhere from a minute to half an hour. OpenAI emphasizes that users should still review and test all AI-generated code before using it in production.

The tool is currently rolling out to ChatGPT Pro, Enterprise, and Team users at no extra cost for a limited time. Support for Plus and Edu plans will follow. OpenAI plans to introduce rate limits and a new pricing model.

Codex aims to reduce the burden of routine programming tasks while offering more transparency and reliability than previous AI coding tools.