OpenAI has launched a research preview of Codex, a cloud-based coding AI agent for software engineers. The tool is available to ChatGPT Pro, Team, and Enterprise users, with plans to extend to Plus users soon.
Overview of Codex’s core capabilities
In a blog post, OpenAI said that Codex is designed to simplify coding tasks such as bug fixing, feature writing, and code reviews. It can run the tasks within isolated cloud sandbox environments, preloaded with user repository.
The tool operates on the codex-1 model, an iteration of OpenAI’s o3 model optimised for software engineering. “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 the blog post.
Users can interact with Codex through ChatGPT by assigning tasks via prompts and tracking progress in real-time. Each task is executed in a separate environment preloaded with the user’s codebase, with completion times ranging from one to 30 minutes based on complexity. OpenAI also claimed that Codex has been trained to reject malicious software development requests while supporting legitimate tasks, reinforced by updated policy frameworks and safety evaluations.
Codex is already being utilised by OpenAI’s technical teams to manage repetitive tasks such as refactoring and testing. Feedback from early testers, including companies such as Cisco and Superhuman, has led to recommendations for assigning well-scoped tasks to multiple agents and experimenting with various task prompts.
As a research preview, Codex currently lacks features such as frontend image inputs and real-time task correction. However, OpenAI envisions a future where developers can delegate routine tasks to AI agents.
“Over time, interacting with Codex agents will increasingly resemble asynchronous collaboration with colleagues. As model capabilities advance, we anticipate agents handling more complex tasks over extended periods,” OpenAI said.
Additionally, OpenAI has launched Codex CLI, an open-source coding agent for terminal use, integrating models like o3 and o4-mini into local workflows. The latest release includes a smaller version of codex-1, optimised for low-latency code Q&A and editing.
Codex joins competitive AI tool marketplace
The introduction of Codex comes amid a surge in AI tools for software engineers. It is poised to compete with Microsoft’s Copilot, an AI-powered coding assistant within Visual Studio and Visual Studio Code. In recent months, Anthropic released agentic coding tool Claude Code while Google enhanced its AI coding assistant Gemini Code Assist.
OpenAI forayed into this space amid reports that the ChatGPT maker is acquiring AI-assisted coding platform Windsurf for $3bn.
Read more: OpenAI expands ChatGPT capabilities with launch of o3 and o4-mini models
