OpenAI is moving deeper into software development with Codex, a new AI coding agent launching in ChatGPT as a research preview. The tool is designed to write code, fix bugs, answer questions about codebases, and run tests from a sandboxed cloud environment.
The launch gives ChatGPT subscribers a more specialized way to use AI for engineering work. It also puts OpenAI more directly into a fast-growing market where AI coding tools have become a major focus for developers, startups, and large technology companies.
What Codex does inside ChatGPT
Codex is powered by codex-1, a version of OpenAI's o3 AI reasoning model that has been optimized for software engineering tasks. OpenAI says codex-1 produces "cleaner" code than o3, follows instructions more precisely, and can keep running tests on its own code until it reaches passing results.
The agent runs in a sandboxed, virtual computer in the cloud. When a user connects GitHub, that environment can be loaded with the user's code repositories, giving Codex the context it needs to work on real projects rather than isolated examples.
OpenAI says Codex can take anywhere from one to 30 minutes to complete simple features, repair bugs, answer questions about a codebase, run tests, and perform related software engineering tasks. The important distinction is that Codex is not only a chat assistant producing snippets. It is presented as an agent that can be assigned work and then execute that work in its own environment.
Users with access can find Codex in the ChatGPT sidebar. They can type a prompt and click the "Code" button to assign a coding task, or ask questions about their codebase and click the "Ask" button. The interface also shows other tasks assigned to Codex and lets users monitor progress.
Why parallel work matters
OpenAI says Codex can handle multiple software engineering tasks simultaneously. That matters because developer work often contains separate threads: one bug fix, one test run, one documentation draft, or one small feature can move independently from another.
The company also says Codex does not block users from using their own computer and browser while the agent is running. In practice, that means the tool is framed as a background worker inside ChatGPT, not a session that demands full attention until it finishes.
In a briefing before launch, OpenAI's Agents Research Lead, Josh Tobin, told TechCrunch the long-term goal is for AI coding agents to become "virtual teammates" that can complete tasks autonomously that take human engineers "hours or even days" to accomplish. OpenAI also says it already uses Codex internally to offload repetitive work, scaffold new features, and draft documentation.
Who gets access first
Codex is rolling out first to ChatGPT Pro, Enterprise, and Team subscribers. OpenAI says those users will start with "generous access," but rate limits are expected in the coming weeks.
After those limits arrive, users will have the option to buy additional credits for Codex, an OpenAI spokesperson tells TechCrunch. OpenAI also plans to expand access to ChatGPT Plus and Edu users soon.
That rollout strategy places Codex among the growing list of subscriber benefits inside ChatGPT. In the past year, OpenAI has added priority access to Sora, its AI video platform, along with Deep Research and Operator. Codex adds a more developer-focused product to that subscription bundle.
The wider AI coding race
AI tools for software engineers, also known as vibe coders, have grown quickly in recent months. The CEOs of Google and Microsoft claim that roughly 30% of their companies' code is now written by AI.
Other companies have also pushed into agentic coding. Anthropic released Claude Code in February, and Google updated Gemini Code Assist in April with more agentic abilities.
The business momentum has been significant. Cursor, one of the most popular AI coding tools, reached annualized revenue of around $300 million in April and is reportedly raising new funds at a $9 billion valuation. OpenAI has also reportedly closed on a deal to acquire Windsurf, the developer behind another popular AI coding platform, for $3 billion.
Codex therefore arrives as both a product launch and a strategic signal. OpenAI is not only adding coding features around ChatGPT; it is building a dedicated AI coding tool at a time when demand for these products is rising across the software industry.
Safety limits and open questions
OpenAI Product Lead Alexander Embiricos says much of the safety work for o3 also applies to Codex. In a blog post, OpenAI says Codex will reliably refuse requests to develop "malicious software."
Codex also operates in an air-gapped environment, without access to the broader internet or external APIs. That design can reduce the danger of misuse, but it may also limit what the agent can do when a task depends on outside services or current online information.
There is another practical issue: AI coding agents can still make mistakes. TechCrunch cites a recent study from Microsoft finding that leading AI coding models, including Claude 3.7 Sonnet and o3-mini, struggled to reliably debug software. That limitation has not stopped investor interest, but it does mean teams still need to treat AI-generated code as work that requires review.
Alongside Codex in ChatGPT, OpenAI is updating Codex CLI, its open source coding agent that runs in the terminal. The company is making a software-engineering-optimized version of o4-mini the default in Codex CLI, and the model will be available through OpenAI's API for $1.50 per 1M input tokens and $6 per 1M output tokens.
The bigger picture is clear: OpenAI wants ChatGPT to be more than a chatbot. With Codex, it is betting that developers will use ChatGPT not just to explain code, but to assign coding work, track progress, and pay for deeper access when the agent becomes useful enough to hit its limits.