OpenAI Codex

Overview

OpenAI Codex is OpenAI's software engineering agent, relaunched in May 2025 (research preview announced May 16, 2025) as a cloud-based agent that works on coding tasks in parallel rather than only completing code. It is distinct from the original 2021 Codex model: this Codex reads a repository, makes changes, runs tests, and proposes work for review. The original cloud preview ran on codex-1, described by OpenAI as a version of its o3 reasoning model optimized for software engineering, and the product has since shipped on later Codex-tuned models.

What it does

Given a prompt and a codebase, Codex writes features, fixes bugs, answers questions about the code, runs and iterates on tests, refactors, and handles migrations. Cloud tasks run inside isolated, sandboxed environments preloaded with the user's repo; Codex reads files, edits code, and runs commands, returning diffs and command/test logs for inspection. It can open changes as GitHub pull requests for human review. It also reviews code, flagging potential bugs, logic errors, and unhandled edge cases.

Autonomy note

Codex is a supervised agent. It performs multi-step work autonomously inside its sandbox but surfaces diffs, logs, and pull requests for a developer to approve before changes land. In the CLI, it runs under configurable sandbox modes (read-only, workspace-write, full-access) and approval policies, so the human controls how far it can act. Treat OpenAI's adoption and benchmark figures as vendor-reported.

Integrations & setup

Codex is available as Codex Web (chatgpt.com/codex), the open-source Codex CLI in the terminal, IDE extensions for VS Code, JetBrains, and Xcode, a desktop app for Windows and macOS, and integrations with GitHub, Slack, and Linear. The CLI supports MCP (Model Context Protocol) for tool integration and is configured via a local config file (sandbox mode, approval policy, and similar settings).

Pricing

Codex is included in paid ChatGPT plans rather than sold separately: Free ($0), Go ($8/mo), Plus ($20/mo), and Pro (from $100/mo, with 5x and 20x rate-limit tiers), plus Business, Edu, and Enterprise options. Cloud usage is metered per rolling 5-hour window and scales by plan. The CLI can alternatively run against an OpenAI API key billed at standard per-token API rates.

Best for / not for

Best for developers and teams who want one OpenAI coding agent across the terminal, IDE, cloud, and GitHub, especially for parallelizable tasks (bug fixes, refactors, test passes) reviewed via diffs and PRs. Less suited to fully unattended automation where no human reviews the agent's output, or to teams that prefer a model-agnostic tool (Codex is tied to OpenAI models).

Alternatives

Claude Code, Cursor, GitHub Copilot, Cognition's Devin, Amazon Q Developer, and Aider are the main competing coding agents and assistants. Codex differentiates on being OpenAI's first-party agent spanning cloud, CLI, IDE, and GitHub from a single ChatGPT account.