OpenAI launches Codex, an AI coding agent, in ChatGPT

OpenAi announced on Friday that it is launching a research example of Codex, the most capable AI coding agent in the company so far.

Codex is powered by Codex-1, a version of the company’s O3 AI-reasoning model optimized for software engineering tasks. OpenAI says that Codex-1 produces the “cleaner” code than O3, more accurately adheres to instructions, and will perform repeat tests on its code until the passing results are achieved.

The Codex agent runs in a sandboxed, virtual computer in the cloud. By making contact with Github, the Codex area can be loaded in advance with your code repositories. OpenAI says that the AI ​​coding agent needs one to 30 minutes to write simple functions, to repair bugs, answer questions about your codebase and perform tests, among other tasks.

Codex can simultaneously process multiple software engineering tasks, says OpenAI, and it does not limit users to access their computer and browser while it is active.

Codex will be rolled out today for Subscribers from Chatgpt Pro, Enterprise and Team. OpenAI says that users have “generous access” to Codex to start with, but in the coming weeks the company will implement tariff limits for the tool. Users then have the option to buy extra credits to use Codex, an OpenAI spokesperson tells WAN.

OpenAI is planning to expand Codex access to Chatgpt Plus and EDU users soon.

AI tools for software, also known as vibe coders, have risen more popular in recent months. The CEOs of Google And Microsoft claims that approximately 30% of their companies are now written by AI. In February Anthropic has his own agentic coding tool, Claude Code, and in April, Google has its AI coding assistant, Gemini Code Assist, updated with more agent skills.

All that atmosphere coding has made the companies behind AI code platforms one of the fastest growing technology. Cursor, one of the most popular AI coding tools, reached a turnover of around $ 300 million in April and is reportedly raised new funds with a rating of $ 9 billion.

Now OpenAi wants a piece of the cake. The chatgpt -maker is reportedly closed on a deal to acquire windsurfThe developer behind another popular AI coding platform, for $ 3 billion. The launch of Codex shows very clearly that OpenAi also builds up its own AI coding tools.

Users with access to Codex can find the tool in the sidebar of Chatgpt and the agent allocated new coding tasks by typing a prompt and clicking the “code” button. Users can also ask questions about their codebase and click on the “Question” button. Under the designation bar, users can see other tasks for which they have assigned Codex and monitor their progress.

In a briefing prior to the launch of Codex, Josh Tinbin of OpenAi’s Agents Research, WAN told that the company ultimately wants his AI coding agents to act such as ‘virtual teammates’, those tasks autonomously perform those ‘hours or even days’ to reach. OpenAI claims that the Codex already uses internally to discharge repetitive tasks, to load new functions and concept documentation.

OpenAi product leader Alexander Embiricos says that much of the safety work for the O3 model of the company also applies to Codex. In a blog post, OpenAI says that Codex reliable will refuse to develop ‘malignant software’ in reliable way. In addition, Codex works in an air walk, without access to the wider internet or external APIs. This limits how dangerous codex can be in the hands of a bad actor – but it can also hinder its usefulness.

It is worth noting that AI coding agents, just like all generative AI systems today, are susceptible to errors. A recent study by Microsoft showed that leading AI coding models, such as Claude 3.7 Sonnet and O3-Mini, struggled to reliably debug software. However, that does not seem to dampen the excitement of investors in these tools.

OpenAI also updates Codex CLI, the recently launched open source coding agent launched in your terminal, with a version of the O4-mini model that is optimized for software engineering. That model is now the standard setting in Codex CLI and will be available in the API of OpenAI for $ 1.50 per 1m input tokens (about 750,000 words, more than the entire Lord of the Rings book series) and $ 6 per 1m export tokens.

Codex’s launch marks the latest effort from OpenAi to strengthen chatgpt with extra products in addition to the infamous chatbot. In the past year, OpenAI has added priority access to the AI ​​video platform of the company, Sora, the research agent, Deep Research, as well as her web browsing agent, operator, as benefits for subscribers.

These offers can seduce more users to register for a Chatgpt subscription and, in the case of Codex, specifically convince existing subscribers to pay OpenAi more money for increased tariff limits.