OpenAI has announced a preview of Codex, an intelligent assistant for programmers designed to optimize the software development process.
Codex is based on codex-1, a version of the o3 reasoning model optimized for software development tasks. OpenAI claims that codex-1 generates "cleaner" code than o3, follows instructions more closely, and will rerun tests of its code until it passes. The new model also supports integration with GitHub for storing code repositories.
Codex can perform multiple tasks simultaneously, operating in a secure cloud environment that prevents data leakage and external interference. Users can grant it access to their codebases and issue commands via text prompts. The new AI assistant is able to work autonomously from 1 to 30 minutes, analyzing the code and suggesting optimal solutions. Its work is displayed in real time, and after the task is completed, Codex updates the repositories and provides detailed reports on the changes made.
OpenAI claims that Codex has built-in security mechanisms to prevent the creation of malware. However, like any AI model, it is not perfect and can make mistakes.
An OpenAI spokesperson told TechCrunch that users will initially receive “generous access” to Codex, but the company will introduce a rate cap on the tool in a few weeks, after which users will be able to purchase additional credits to use Codex.
Codex is currently available to ChatGPT Pro, Enterprise, and Team users, and will be available to Plus and Edu subscribers in the future.