OpenAI has launched a cloud-based software engineering AI agent powered by its most advanced coding model, ‘codex-1’. Starting today, Codex is available to ChatGPT Pro, Team, and Enterprise users, with access for Plus and Edu users coming soon.
Codex is designed to handle multiple tasks in parallel in the cloud. It can add features, fix bugs, answer questions about your codebase, and even generate pull requests for review. Developers can link their GitHub repositories and use the AI agent to streamline their workflows. Codex can be accessed directly from the left sidebar in ChatGPT.
OpenAI states that codex-1 is built on the o3 model and trained “using reinforcement learning on real-world coding tasks across various environments. This approach enables it to generate code that closely aligns with human coding styles and pull request (PR) preferences.“
Codex carefully follows instructions and runs tests repeatedly in a cloud environment until it gets a passing result. You can include an AGENTS.md file in your repository to guide the agent, helping it understand your codebase and perform tasks exactly as you intend.
On OpenAI’s internal software engineering benchmarks, codex-1 achieves 75% accuracy—outperforming the o3-high model, which scores 70%. It also slightly edges out o3-high on the SWE-Bench Verified benchmark after multiple attempts.
OpenAI also emphasizes that Codex operates within a secure, isolated cloud container. During task execution, the agent has no internet access, meaning it cannot connect to external websites, APIs, or services.
