OpenAI introduced Codex, the most powerful AI agent for writing code in ChatGPT

OpenAI announced the launch of the preview version of Codex — the company’s most powerful AI agent for programming. Codex is based on the codex-1 model, optimized for engineering tasks, creating “clean” code and automatically testing it until successful.  

Codex operates in a cloud virtual environment, connects to GitHub, and can load user repositories. It can write code, fix errors, and run tests within 1–30 minutes.  

Codex is already available to ChatGPT Pro, Enterprise, and Team subscribers, and will soon be part of ChatGPT Plus and Edu. OpenAI plans to limit access to the tool, offering users the option to purchase additional credits.  

The popularity of AI coding is growing rapidly: Microsoft and Google state that about 30% of their code is now written by AI. OpenAI is also actively developing this area, recently acquiring Windsurf for $3 billion.  

Codex can be found in the ChatGPT sidebar, where users can assign tasks and track their progress. OpenAI views Codex as a “virtual partner” capable of autonomously performing complex engineering tasks.  

To enhance security, Codex operates in an isolated environment without internet access or external APIs, preventing the creation of malicious software. Despite potential errors, investor interest in AI coding remains high.  

In addition to Codex, OpenAI is updating Codex CLI — an open-source AI agent for the terminal using the optimized o4-mini model. The launch of Codex is another step by OpenAI to expand ChatGPT’s capabilities, aimed at attracting new subscribers and increasing revenue.

Source: TechCrunch