OpenAI has introduced Codex, a cloud-based AI assistant that helps developers automate routine tasks. It can write code, fix errors, run tests, and suggest changes to repositories. It is reported by the company’s press service.
Codex is built into the web version of ChatGPT. It works on the basis of a specialized version of OpenAI’s o3 model called codex-1, which is specially optimized for programming tasks. OpenAI also claims that Codex will complement other AI tools such as Cursor and Windsurf code editor.
The peculiarity of Codex is that it performs each task in a separate “virtual space” that already has everything necessary to work with the code.
It supports the simultaneous execution of several tasks, and the execution time depends on the complexity and can range from 1 to 30 minutes. You can also integrate Codex with your GitHub account.
Codex is based on a special version of artificial intelligence from OpenAI that has been trained on a large number of real-world programming tasks:
- Imitates human coding style.
- Follows instructions precisely.
- Can iteratively run tests until it achieves successful results.
- Codex generates code that is very similar to the way people write, follows instructions precisely, and can check its own work until it gets the right result.
In addition, Codex provides a detailed report of each of its steps, showing exactly what it did to complete the task.