OpenAI Introduces Codex AI Agent to Automate Software Engineering in ChatGPT
OpenAI launches Codex, a cloud-based AI coding agent within ChatGPT that autonomously writes features and fixes bugs. The tool is initially available to Pro, Enterprise, and Team subscribers using the new codex-1 model.
OpenAI launches Codex, a new AI coding agent inside ChatGPT that autonomously handles software engineering tasks. Powered by a specialized model called codex-1, which is an optimized version of the o3 reasoning model, the tool writes features, fixes bugs, and answers questions about codebases. OpenAI states that codex-1 produces cleaner code and adheres more precisely to user instructions than its predecessor.
The Codex agent operates within a sandboxed, cloud-based virtual computer rather than taking over a user's local machine. By connecting directly with GitHub, the environment comes preloaded with the user's existing code repositories. Codex runs tests iteratively until the code passes, taking anywhere from one to thirty minutes to complete a task while allowing users to continue using their own computers in the meantime.
Access to Codex begins today for ChatGPT Pro, Enterprise, and Team subscribers, with plans to expand to Plus and Edu users soon. OpenAI offers generous access initially but intends to implement rate limits in the coming weeks. Users will then have the option to purchase additional credits to continue using the autonomous coding tool.