“It’s not just a tool — it’s a collaborator, an intern, a mentor, and a team, all in one.”
At OpenAI, we’ve long believed that software engineering is ripe for transformation. In 2021, we introduced Codex — an early model that pushed beyond autocomplete and hinted at a new era of AI-assisted development. That journey continued with the evolution of models like O1 and O1 Mini, which laid the groundwork for scalable AI coding tools. That vision has now culminated in something far more powerful: OpenAI Codex — a new standard for the OpenAI software agent.
Today, we’re launching Codex by OpenAI, a next-generation, remote AI software agent designed to fundamentally change how software is built, tested, and maintained.
What Is OpenAI Codex?
OpenAI Codex is an intelligent, autonomous software agent that operates in the cloud on OpenAI’s infrastructure. It brings AI-native engineering workflows directly into your repositories and can:
- Analyze and explain codebases
- Identify and fix bugs
- Propose proactive improvements
- Run linters and formatting tools
- Write and test code — even create pull requests
This OpenAI software agent operates asynchronously, in parallel, and at scale — so whether you want to tackle one bug or spin up a hundred concurrent tasks, OpenAI Codex is built for it. Think of it as your own scalable AI development team.
Why OpenAI Codex Is Different
The foundation of OpenAI Codex is our latest model, Codex One — built not just for performance, but for practicality. It’s optimized for real-world code that’s readable, mergeable, and production-ready. This builds on the advancements seen in smaller, efficient models like ChatGPT O3 Mini, which demonstrated how compact models can still deliver powerful coding capabilities.
Codex by OpenAI goes beyond static suggestions. It integrates deeply into your development environment:
- Reads your repo, environment variables, secrets, and dev setup
- Runs in secure, sandboxed microVMs
- Uses built-in tools for linting, testing, debugging, and formatting
This isn’t just a code generator — it’s a fully capable OpenAI software agent you can trust.
Built for End-to-End Autonomy
Trained through end-to-end reinforcement learning, OpenAI Codex understands how to:
- Navigate complex project structures
- Reproduce vague bug reports
- Apply contextual code changes
- Run and interpret test suites
- Write thoughtful commit messages and follow repo conventions
In internal testing, Codex by OpenAI autonomously:
- Found and fixed real bugs
- Wrote and passed tests
- Prepared PRs ready for human review
This is true agentic development — not just autocomplete with extra steps.
Customizable and Powerful
Each OpenAI Codex task runs in a dedicated, configurable environment. Developers can preload linters, formatters, test dependencies, and more.
Want to guide the agent? Add an agents.md
file to your repo. From commit message styles to ASCII-art easter eggs, Codex by OpenAI listens and adapts to your development style.
Verifiability You Can Trust
Transparency is a core feature of OpenAI Codex. Every task includes:
- A summary of what it did and why
- Citations for every line of code referenced
- Logs of executed tests and their results
This makes Codex by OpenAI one of the most verifiable and trustworthy AI software agents available — often more transparent than human collaborators.
Real Use Cases from OpenAI
Within OpenAI, we’ve already used OpenAI Codex to:
- Refactor large-scale codebases
- Automatically resolve CI failures
- Propose meaningful engineering improvements
- Reduce code review burdens with clean, production-ready PRs
The results are transformational. Tasks that once took hours now take minutes — or run entirely in the background via a scalable OpenAI software agent.
Local + Cloud: One Seamless System
OpenAI Codex is available in two modes:
- Codex CLI: A local agent that runs in your terminal — ideal for quick, synchronous development
- Codex Cloud: A powerful remote OpenAI software agent that runs tasks in parallel and at scale, on OpenAI compute
They’re designed to work together. Start a task locally, let the cloud agent finish it while you step away, and return to a fully tested PR — without touching your local machine.
Now Available for Pro and Enterprise Users
Starting today, OpenAI Codex is available to:
- ChatGPT Pro
- ChatGPT Enterprise
- ChatGPT Teams
Support for ChatGPT Plus and Edu is coming soon. We’re also rolling out:
- CI/CD and issue tracker integration
- API access for task automation
- Seamless scaling with generous usage limits
There’s no additional pricing at launch.
Let’s Build the Future — Together
We created OpenAI Codex to supercharge developers everywhere — from indie hackers to global engineering teams. It’s your AI-powered collaborator, pair programmer, and code maintainer — all in one.
The way we build software is changing. It’s asynchronous. It’s autonomous.
It’s powered by Codex by OpenAI — and it’s here.
Try OpenAI Codex today and be part of the next generation of software engineering.
Explore the full blog and demo videos →
Comments