OpenAI debuts Codex AI Coding Agent in ChatGPT
OpenAI has introduced Codex, its most advanced AI coding agent to date, as part of a research preview now available in ChatGPT. Built on codex-1 a version of the o3 AI model tailored for programming Codex is designed to follow instructions more closely, generate cleaner code, and automatically test until the code runs successfully.
Codex operates inside a secure, virtual cloud environment. It integrates with GitHub, allowing it to preload a user’s repositories for faster setup. Users can assign tasks such as writing features, debugging, answering codebase questions, or running tests. While tasks may take between one and 30 minutes, users remain free to use their computers and browsers during execution. Notably, Codex can manage multiple tasks in parallel, making it a flexible assistant for software engineers.
The agent is now rolling out to ChatGPT Pro, Team, and Enterprise users. OpenAI says these users will initially enjoy generous access, though rate limits will be introduced soon. After that, users may purchase extra credits to maintain higher usage levels.
Codex Enters a Competitive AI Coding Market
Codex’s debut reflects OpenAI’s intent to capture a growing share of the AI coding tools market. In recent months, other tech giants have made similar moves. For instance, Google upgraded its Gemini Code Assist in April, while Anthropic released Claude Code earlier this year. According to the CEOs of both Microsoft and Google, nearly 30% of their companies’ code now originates from AI.
Furthermore, OpenAI has reportedly acquired Windsurf, a rival AI coding platform, in a $3 billion deal. These developments highlight the rapid growth of the industry Cursor, another coding tool, recently hit $300 million in annual revenue and may soon be valued at $9 billion.
Clearly, OpenAI aims to strengthen its position not just through product launches but also strategic acquisitions.
AI Coding Agents Take Shape as Virtual Teammates
Codex allows users to assign tasks through ChatGPT’s interface by entering prompts and selecting “Code” or “Ask” buttons. Tasks and their statuses are displayed beneath the input area. OpenAI envisions Codex evolving into a virtual teammate capable of completing tasks in hours that might take humans days. Internally, OpenAI is already using Codex to handle repetitive coding tasks and assist in documentation.
Despite its capabilities, Codex does have limits. It runs in an isolated environment, which blocks internet and external API access. While this boosts safety preventing the generation of harmful code it can also reduce functionality. OpenAI stresses that Codex will refuse to create malicious software, aligning with its broader safety commitments.
It’s important to note that AI coding agents remain imperfect. A Microsoft study showed that leading models still struggle with debugging. However, excitement around these tools continues to grow. OpenAI also updated its Codex CLI an open-source terminal agent with an o4-mini model optimized for coding, now the default. Pricing for API use starts at $1.50 per million input tokens and $6 per million output tokens.
Codex’s launch adds to OpenAI’s expanding suite of tools for ChatGPT subscribers, alongside Sora (AI video), Operator (web browsing), and Deep Research. These offerings aim to attract more users while increasing the platform’s value through advanced, practical features.
