OpenAI Codex is OpenAI's coding agent: a software engineering assistant that can read, plan, write, test, and modify code inside a sandboxed execution environment, shipped as one connected product across ChatGPT (web and desktop), a command-line tool (Codex CLI), an IDE extension, and a cloud/async execution mode. This is the current, actively developed Codex product line, distinct from the original 2021 Codex API (the code-davinci-002-era model that once powered early versions of GitHub Copilot); OpenAI retired API access to that legacy model in 2023, and today "Codex" refers to the agentic coding product described here.
Company
Codex is developed and operated by OpenAI, the company behind ChatGPT and the GPT model family. Inside OpenAI's product line-up, Codex is positioned as the dedicated tool for professional software engineering work, distinct from ChatGPT as a general-purpose chat assistant.
Core features
Codex is built around handing the agent a starting point — a goal, an idea, or a task — and letting it gather context across a codebase, take action, and produce a reviewable result such as a diff, a pull request, or a running change. According to OpenAI's own product documentation, Codex can:
- Work locally through Codex CLI or inside an editor via the Codex IDE extension, or run tasks remotely in Codex cloud, with the ability to hand tasks off between a local machine and a remote host.
- Execute inside a sandbox with configurable agent approvals and security settings, including permission profiles and control over internet access, plus an auto-review step before changes are applied.
- Take actions beyond editing files, including browser and computer-use functionality, image generation, and web search when a task calls for it.
- Turn a demonstrated workflow into a reusable "skill" through a record-and-replay capability.
- Run scheduled or long-running tasks that continue in the background instead of requiring a developer to stay in the loop.
- Use a dedicated security-review capability to scan and flag risky changes.
Pricing and access
Codex is not sold as a separate subscription; access is bundled into ChatGPT plans. Usage is included, up to plan-specific limits, in the Free, Go, Plus, Pro, Business, Edu, and Enterprise tiers, and ChatGPT Business and Codex usage are now billed together on the same seat. Once included usage runs out, further Codex use is metered through a credit system whose cost varies by model, context size, reasoning effort, and tools used. Separately, Codex can be run against an OpenAI API key on a pay-as-you-go basis, charged at standard API token rates rather than through a ChatGPT plan. OpenAI has also introduced flexible, usage-based pricing aimed at teams running many Codex tasks in parallel, on top of seat-based Business and Enterprise plans.
Who it's for
Codex targets professional developers and engineering teams who want an agent to handle real, multi-file coding work — bug fixes, refactors, test-writing, reviewing pull requests, or building small features — rather than just inline autocomplete suggestions. The combination of a local CLI, an IDE extension, and a cloud execution mode suits teams that want to run coding tasks both interactively and asynchronously, including workflows where many tasks run unattended in the background and are reviewed once complete. Business and Enterprise customers get seat-based billing, data-handling controls, and a data processing addendum, positioning Codex for organizations with compliance requirements alongside individual developers.