OpenAI Launches Its Most Advanced AI Coding Agent Yet
Codex allows users to review its work through citations, terminal logs, and test results.
OpenAI has launched a research preview of Codex, its most advanced AI coding agent, powered by the codex-1 model—a version of the o3 reasoning model optimised for software engineering.
Codex produces cleaner code, better follows instructions, and can automatically test code until it passes.
Codex operates in a sandboxed virtual cloud environment and connects with GitHub, enabling access to users' code repositories.
It can write features, fix bugs, run tests, and answer codebase questions in as little as one minute or up to 30 minutes. Users can continue to work on their devices while Codex runs in the background.
"It was trained using reinforcement learning on real-world coding tasks in a variety of environments to generate code that closely mirrors human style and PR preferences, adheres precisely to instructions, and can iteratively run tests until it receives a passing result," OpenAI said in a blog post.
In designing Codex, the company prioritised security and transparency to ensure users can verify the model’s outputs—an increasingly critical safeguard as AI takes on more complex coding tasks independently and safety concerns evolve.
Codex allows users to review its work through citations, terminal logs, and test results.
"When uncertain or faced with test failures, the Codex agent explicitly communicates these issues, enabling users to make informed decisions about how to proceed. It still remains essential for users to manually review and validate all agent-generated code before integration and execution," the startup added.
Initially available to ChatGPT Pro, Enterprise, and Team users, Codex will soon expand to Plus and Edu subscribers.
The launch comes amid a surge in AI coding tools, with companies like Google, Microsoft, and Anthropic releasing their own agents.
Windsurf, a company acquired recently, announced the launch of its first family of AI models, SWE-1, which is designed to optimise the entire software engineering process, far beyond basic code generation.
Comments ()