Codex Review 2026: OpenAI's Coding Agent, Plans and Limits

Codex is OpenAI’s coding agent. Like Claude Code, it works at the level of a task rather than a line: you describe what you want, and it reads the repository, edits files, runs commands and iterates. It reaches you in several places: an open-source command-line tool, an extension for VS Code, Cursor and Windsurf, Codex on the web (tasks run in the cloud), the ChatGPT desktop app, and automatic code review on GitHub. It is included in ChatGPT’s paid plans rather than sold separately.

How this review was written. It is based on OpenAI’s Codex pricing page and the openai/codex repository, both checked 2026-10-26, and on public user reports. We have not run a lab benchmark. Where a capability isn’t stated in those sources (MCP support, a JetBrains integration), our comparison matrix shows it as unverified rather than guessing.

The Short Version

Rating: 4/5 (our editorial assessment of documented capabilities, price and user reports). If you already pay for ChatGPT Plus, Codex costs you nothing extra and is worth trying before buying a second agent subscription. Its strengths are breadth of surfaces, an open-source CLI, and cloud tasks that keep running while you do something else. Its limits: no inline autocomplete, OpenAI models only, and usage caps that are counted per five hours on Plus and Business.

Try Codex

Where Codex runs

• CLI. Open source under the Apache-2.0 license. Sign in with your ChatGPT account to use your plan, or use an API key and pay API rates.
• IDE extension. For VS Code, Cursor and Windsurf, per the repository.
• Web and cloud. Codex on the web runs tasks in the cloud against your repositories, so long tasks continue without your machine.
• Desktop app. Through the ChatGPT desktop app. This is also the only way the Free and Go plans reach Codex, with a single model and “subject to rollout”.
• GitHub code review. Automatic reviews on pull requests; per OpenAI, these count toward your usage limits.

Pricing

Prices from OpenAI’s Codex pricing page, checked 2026-10-26 (see our pricing table for every tool):

• Plus, $20/month: the CLI, the IDE extension and Codex on the web, with usage limits counted per five-hour window (and possibly weekly). Extra usage can be bought as ChatGPT credits.
• Pro, from $100/month ($100, $200 or $500 tiers): no five-hour limit.
• Business, $25 per user per month ($20 billed annually, 2+ users): the same limits as Plus per seat, admin controls, and the option to buy credits.
• Enterprise and Edu: custom pricing.
• API key: no subscription; pay OpenAI’s API rates for what it uses.

OpenAI publishes the five-hour allowances as ranges per model (for example, the smallest model allows far more messages than the largest), because the real number depends on task size. Treat them as estimates rather than a quota you can plan around precisely.

Strengths

• No extra cost for ChatGPT subscribers. Plus users already have it.
• Open-source CLI. You can read, audit or fork the client.
• Cloud tasks. Hand off work that continues when you close your laptop, and check on it later.
• Many surfaces. Terminal, three editors, web, desktop and GitHub.

Limitations

• No inline autocomplete. Like other agents, it doesn’t suggest code as you type.
• OpenAI models only. You can’t point it at another provider’s model.
• Five-hour caps on Plus and Business. Long sessions can hit the limit; Pro removes it.
• Unverified in our sources: MCP support and a JetBrains integration aren’t stated on the pages we checked.

Codex vs. the Competition

Compared to Claude Code, the two are the closest match in this matrix: both are task-level agents with a CLI, editor extensions, a desktop app and the web, both included in their vendor’s subscription. The practical deciding factors are which subscription you already pay for, which models you prefer, and how each plan’s usage limits fit your day. See our Claude Code review.

Compared to Cursor or GitHub Copilot, Codex isn’t an editor or an autocomplete tool. Many developers pair an agent like Codex with one of those for help while typing.

Compared to Aider and Cline, which are open source and let you bring any model, Codex is tied to OpenAI models but comes with plan-based pricing instead of a per-token bill.

Who Should Use Codex

Codex fits if you:

• Already pay for ChatGPT Plus, Pro or Business
• Want an agent that can run tasks in the cloud while you work on something else
• Value an open-source client you can inspect

It’s probably not for you if you:

• Want inline autocomplete (look at Copilot or Cursor)
• Need to choose your model provider (look at Cline or Aider)
• Need a verified MCP or JetBrains integration today

Try Codex