OpenAI announced on Friday the launch of a research preview for Codex — its most advanced AI coding agent to date. Reports Technology News
Codex is powered by codex-1, a specialized version of OpenAI’s o3 AI reasoning model fine-tuned for software engineering. According to OpenAI, codex-1 generates “cleaner” code than the standard o3 model, follows instructions more precisely, and continuously tests its code until all requirements are met.
The Codex agent operates in a secure, cloud-based virtual environment. By integrating with GitHub, it can automatically preload your code repositories, allowing seamless interaction with your existing projects. OpenAI notes that Codex can handle tasks like writing features, fixing bugs, answering questions about your code, and running tests — often completing simple tasks in as little as one to 30 minutes.
Codex is built for multitasking and runs independently in the background, meaning developers can continue using their computer and browser while Codex works.
Starting today, ChatGPT Pro, Enterprise, and Team subscribers will gain access to Codex. OpenAI says users will initially enjoy “generous access,” but rate limits will be introduced in the coming weeks. After that, users will have the option to purchase additional usage credits, according to a spokesperson who spoke with TechCrunch.
AI-powered developer tools — sometimes dubbed “vibe coders” — have seen rapid adoption in recent months. Both Google and Microsoft report that around 30% of their code is now written by AI. In response to this growing trend, Anthropic launched Claude Code in February, and Google rolled out an enhanced version of its Gemini Code Assist in April, adding more autonomous coding capabilities.
