OpenAI launches Codex, a new cloud-based coding assistant that can work on many programming tasks at once. Codex can help you write new features, answer questions about your code, fix bugs, and suggest pull requests for review. Each task runs separately in a secure cloud workspace with your code already loaded.

Powered by Codex-1, a version of OpenAI’s model specially trained for coding, Codex writes code like a human, follows instructions carefully, and runs tests until everything works. It’s now available for ChatGPT Pro, Enterprise, and Team users, with Plus and Edu users coming soon.

How does Codex work?

You can use Codex from the ChatGPT sidebar. Just type what you want it to do and click “Code” for coding tasks or “Ask” for questions about your code. Each task runs in its own safe workspace. Codex can read and edit files and run tests or checks. Tasks take from 1 to 30 minutes, depending on their difficulty, and you can watch the progress live.

When Codex finishes, it saves the changes and shows proof of its work with logs and test results. You can review the results, ask for changes, create a GitHub pull request, or add the changes directly to your environment. You can also set up Codex’s workspace to match your own development setup closely.

Codex can be guided by AGENTS.md files placed within your repository. These are text files, akin to README.md, where you can inform Codex how to navigate your codebase, which commands to run for testing, and how best to adhere to your project's standard practices. Like human developers, Codex agents perform best when provided with configured dev environments, reliable testing setups, and clear documentation. 

On coding evaluations and internal benchmarks, Codex-1 shows strong performance even without AGENTS.md files or custom scaffolding.

Codex Availability, Pricing, and Limitations (Simplified)

Starting today, Codex is available in ChatGPT for Pro, Enterprise, and Team users around the world. Support for Plus and Edu users will be added soon. For the next few weeks, users can try Codex for free, after which usage will be limited, with flexible pricing options for extra use.

If you're a developer using the codex-mini-latest model.

• Input tokens cost $1.50 per million
• Output tokens cost $6 per million
• There’s a 75% discount for cached prompts

Codex is still a preview, so some features are missing.

• No image input support for frontend tasks yet
• You can't change what the agent is doing mid-task
• Remote agent responses may feel slower than live editing

But over time, Codex will feel more like working with a helpful teammate who can take on bigger jobs.

What’s Coming Next?

We’re building Codex to help developers work faster by,

• Letting you handle important tasks and delegate the rest to AI
• Supporting both real-time coding help and asynchronous task handling

In the future,

• You'll be able to give instructions during a task
• Codex will offer progress updates and implementation ideas
• It will work smoothly with tools like GitHub, ChatGPT Desktop, and CI systems

We believe this is just the beginning of a major shift in how software is built, with AI helping individuals and teams become more productive.