On May 16, 2025, OpenAI launched a research preview of Codex, a cloud-based software-engineering agent. The name reused OpenAI’s 2021 code model, but the 2025 Codex is a different kind of thing: an agent that can take on many coding tasks in parallel, each running in its own isolated cloud sandbox preloaded with the user’s repository. It can write features, answer questions about a codebase, fix bugs, and propose pull requests for a human to review.
Codex is powered by codex-1, a version of OpenAI’s o3 reasoning model tuned for software engineering through reinforcement learning on real coding tasks, with the goal of producing code and pull requests that match human style and review preferences. For safety the agent runs in a locked-down container with internet access disabled during a task, so it can only touch the code and dependencies the user has explicitly provided. It launched for ChatGPT Pro, Business, and Enterprise users, with Plus access following in June 2025.
Codex matters as a marker of how fast agentic coding moved from demo to product. Where the 2021 Codex powered inline suggestions in an editor, the 2025 Codex assigns work to an autonomous agent in the cloud and waits for a pull request, an interaction model that started to resemble delegating to a junior engineer. For a business reader, it signals a shift in how software teams might allocate routine engineering work.