OpenAI Codex Review 2026: Pricing, Limits & Features

Disclosure: CodeToolPilot may earn a commission if you buy through links on this page, at no extra cost to you. How we make money.

Last verified: October 12, 2026. This is an editorial research review — how we review. Prices and plans change often; confirm on the vendor’s site.

Status update: On July 9, 2026, OpenAI moved the Codex desktop app into the ChatGPT desktop app, alongside ChatGPT Work. The Codex CLI, IDE extension and Codex Cloud continue, with docs now at learn.chatgpt.com.

OpenAI Codex is OpenAI’s software engineering agent. In October 2026 it is less a single app than a family of surfaces tied to your ChatGPT account. There is an open-source command-line tool, an extension for VS Code-compatible editors, integrations for JetBrains IDEs and Xcode, and Codex Cloud for tasks that run on OpenAI’s infrastructure. A Codex experience also lives inside the ChatGPT desktop, web and mobile apps.

Codex is aimed at developers who want to hand off real engineering work: bug hunts, refactors, new features and pull request reviews. You can drive it interactively in a terminal or run several tasks in parallel in the cloud. Every ChatGPT plan includes it, from Free and the $8 Go tier upward, which makes it one of the cheapest ways to try an agentic coding workflow. The lower tiers come with tight limits and a smaller model, though.

This is an editorial research review based on OpenAI’s documentation, pricing and changelog, the Codex GitHub repository and public user feedback. We did not run our own benchmarks.

Quick verdict

Codex is a strong, fast-moving choice for developers already in the ChatGPT ecosystem. The CLI is open source under Apache-2.0, the sandbox and approval model is well documented, and Codex Cloud makes parallel work straightforward, even with your laptop closed. The main caveats are hard-to-predict usage limits on subscription plans and frequent model and feature churn. Codex also now shares its home, docs and branding with ChatGPT’s general-purpose agent, which can make it harder to know exactly what you are buying.

Pro: Included in every ChatGPT plan, from Free and Go up to Pro, Business and Enterprise, and usable with an API key.
Pro: Open-source CLI (Apache-2.0) with documented sandbox modes and approval policies. Network access is off by default for local runs.
Pro: Codex Cloud runs tasks in parallel, each in its own workspace, and can open pull requests.
Pro: Broad editor coverage: VS Code, Cursor, Windsurf, JetBrains IDEs and Xcode.
Con: Allowances are published as wide ranges, such as 15–160 local GPT-6.1 Sol messages per five hours on Plus. Weekly limits are unstated, and users publicly complain about fast-draining quotas.
Con: Rapid churn: GPT-5.5 retires from Codex on October 14, 2026, and the codex mcp-server command was removed in September, less than two weeks after it was deprecated.
Con: Built around OpenAI models, and cloud features are unavailable when you sign in with an API key.
Con: The IDE extension is not open source, and the merge with ChatGPT’s apps blurs the product’s boundaries.

Visit OpenAI Codex

What is OpenAI Codex?

Codex is the umbrella name for OpenAI’s coding agent across several clients. On July 9, 2026, OpenAI moved the Codex desktop experience into the ChatGPT desktop app, alongside Chat and ChatGPT Work, a general-purpose agent built on Codex. Codex documentation now lives on learn.chatgpt.com, next to ChatGPT’s own docs. OpenAI’s product page still presents Codex as a coding product that you use “across ChatGPT, your editor, and the terminal, all connected by your ChatGPT account.”

OpenAI lists three components as open source in the openai/codex GitHub repository: the CLI, the Codex SDK and the app server. The repository is licensed under Apache-2.0. The IDE extension is listed as not open source. The CLI installs via npm, Homebrew, a standalone script or prebuilt binaries. OpenAI says Codex runs natively on Windows through the desktop app, the CLI and the IDE extension, as well as on macOS, Linux and WSL2.

Key features

CLI and IDE extension

The CLI is an interactive terminal agent. /permissions sets what Codex may do without asking, and /init writes an AGENTS.md file of project instructions. codex resume reopens recent chats, and codex exec runs non-interactively for scripts and CI. The IDE extension works in Visual Studio Code, VS Code Insiders, Cursor and Windsurf. JetBrains IDEs (through AI Chat) and Xcode have their own integrations. Inside the editor you can add open files or selections as context, review changed lines in place, create git checkpoints before and after a task, or hand a longer job to the cloud.

Codex Cloud

Codex Cloud runs tasks remotely from a published environment. Each task gets its own workspace, keeps working while your computer sleeps, and can end in a commit or pull request. Codex can install dependencies and walk through setup, and an environment setting controls whether it may reach the internet. GitHub is the main integration, and GitLab support for Codex Cloud entered beta on August 19, 2026. You can follow and steer tasks from the web, mobile or desktop apps.

Sandboxing and approvals

Three sandbox modes control what local commands can touch. read-only is for chat and planning. workspace-write is the default in version-controlled folders and allows edits inside the workspace. danger-full-access removes restrictions. Approval policies (on-request, never and a granular option) decide when Codex asks before acting. Network access is off by default. On Windows, Codex offers native sandbox implementations, with WSL2 using the Linux sandbox.

Models and speed modes

OpenAI recommends GPT-6.1 Sol, added on September 29, 2026, for complex coding and agentic work. GPT-6 Astra is listed as the most capable model, and GPT-6 Luna is the efficient option for focused, high-volume tasks. Free and Go users get Luna. A Fast mode uses 2.5x your included usage, and Ultrafast uses 8x. Ultrafast is available only on the $500 Pro tier and eligible Enterprise and Edu plans.

Code review, subagents and extensions

Codex can review pull requests automatically and delegate focused work to subagents. It supports local and remote MCP servers, plugins and skills. Since August 25, 2026, eligible plans can trigger scheduled tasks from Gmail, Slack and GitHub events. OpenAI also lists integrations for GitHub (an Action and an app), Slack, Linear and GitLab (beta).

Migration from other agents

Since August 11, 2026, the CLI’s /import command can bring over supported setups from Claude Code and Cursor, which lowers the cost of trying Codex.

Who should use OpenAI Codex

Codex makes the most sense for developers who already pay for ChatGPT Plus, Pro or Business and want an agent in the terminal and editor without another subscription. It also suits teams that want open-source CLI tooling with a documented sandbox, and anyone who wants to run batches of cloud tasks.

Who should skip it

Skip Codex if you need predictable, published quotas or rely on stable model names and CLI commands in automation. It is also a poor fit if you want to use Anthropic, Google or open-weight models in the same agent, or if your organization cannot use ChatGPT accounts and needs cloud features.

Strengths and limitations

Strengths

Low barrier to entry. Free and $8 tiers make it easy to try, with one account across terminal, editor, web and phone.
Open-source core. The CLI, SDK and app server can be audited and forked, which helps security reviews.
Clear safety model. Named sandbox modes, network off by default and native Windows sandboxing are documented in detail.
Parallel cloud work. Codex Cloud suits batches of independent tasks reviewed later as pull requests.
Model and speed choices. OpenAI says GPT-6 Sol and Luna are rolling out at lower token prices than GPT-5.6, while GPT-6 Astra is the costlier top tier. Fast and Ultrafast modes trade allowance for speed.

Limitations

Unpredictable allowances. OpenAI publishes ranges rather than fixed numbers, says weekly limits “may also apply” without stating them, and notes that cloud tasks cost more. A long-running OpenAI Developer Community thread started on August 14, 2026 collects subscriber complaints about weekly limits draining much faster than before.
Frequent breaking changes. GPT-5.3-Codex-Spark was deprecated on September 14, 2026, and GPT-5.5 retires from Codex on October 14, 2026. The codex mcp-server command was deprecated on August 24 and removed on September 5. The old untrusted approval setting is retired and can prevent startup if left in a config file.
Blurred product boundaries. Codex, ChatGPT Work and the ChatGPT apps now share docs, limits and a desktop app, so it takes effort to know which features and quotas apply to coding.
Tied to OpenAI and a ChatGPT sign-in. Codex is designed around OpenAI’s models. Codex Cloud requires a ChatGPT sign-in, and Enterprise and Edu admins must enable GPT-6.1 Sol before members can use it.
Partly closed. The IDE extension is not open source, even though the CLI is.

Pricing (as of October 2026)

Plan	Price	What you get
Free	$0	Codex for quick coding tasks. GPT-6 Luna at standard speed in the desktop app, subject to rollout.
Go	$8/month	Codex for lightweight coding tasks. GPT-6.1 Sol was not included at launch.
Plus	$20/month	“A few focused coding sessions each week.” Estimated local messages per five hours: GPT-6.1 Sol 15–160, GPT-6 Astra 5–45, GPT-6 Luna 350–3,000. Weekly limits may also apply.
Pro	$100, $200 or $500/month	No five-hour limit at present, though weekly limits may apply. Ultrafast speed on the $500 tier.
Business	$20/user/month billed annually (2+ users), or $25/user/month monthly	Standard Business seats have the same estimated limits as Plus, plus workspace credits and admin controls.
Enterprise and Edu	Contact sales	Flexible pricing with no fixed rate limits. GPT-6.1 Sol stays off until an admin enables it.
API key	OpenAI API token pricing	CLI, SDK and IDE extension only, with no cloud features.

Cloud tasks can use more of your allowance than local messages. Plus and Pro users can buy extra credits. Confirm current prices and limits on OpenAI’s Codex pricing page, since OpenAI has been changing models and allowances frequently.

See OpenAI Codex plans

Prices verified on October 12, 2026 against the official OpenAI Codex pricing page. Plans change often — confirm on the vendor’s page before you buy.

Use cases

OpenAI Codex spans a terminal agent, editor integrations and cloud tasks, so it fits several kinds of engineering work. These scenarios are based on OpenAI’s documentation and the research behind our review, not on hands-on testing.

Clearing a backlog of independent tasks in the cloud

A developer with a list of small, well-defined issues wants them worked on in parallel and returned as pull requests. Codex Cloud gives each task its own workspace, keeps working while your computer sleeps, and can finish with a commit or pull request, with GitHub as the main integration and GitLab support in beta. The caveats are that cloud tasks can use more of your allowance than local messages, and cloud features require a ChatGPT sign-in rather than an API key.

Sandboxed refactoring in the terminal

A developer wants an agent to refactor code locally without giving it free rein over the machine. The open-source CLI defaults to the workspace-write sandbox in version-controlled folders, keeps network access off by default, and lets you set what it may do without asking through /permissions, while /init writes an AGENTS.md file of project instructions. Avoid danger-full-access mode unless you accept that it removes these restrictions.

Scripted runs in CI

A platform team wants to run an agent inside scripts or CI jobs. codex exec runs Codex non-interactively, and OpenAI also lists a GitHub Action among its integrations. Because OpenAI retires models and changes CLI commands often, as with the removal of codex mcp-server in September 2026, we would pin versions and recheck configs after each update.

Automatic pull request reviews and event-driven tasks

A team wants an extra reviewer on every pull request and some routine work triggered by events. Codex can review pull requests automatically, and since August 25, 2026, eligible plans can trigger scheduled tasks from Gmail, Slack and GitHub events. Check which plans are eligible before planning a rollout around these features.

Trying agentic coding, or switching from another agent

A developer who already has a ChatGPT account wants to try agentic coding before paying for another tool, or to move an existing setup from Claude Code or Cursor. Codex is included in every ChatGPT plan, including Free and Go, and the CLI’s /import command brings over supported setups from Claude Code and Cursor. Free and Go users get the smaller GPT-6 Luna model with tight limits, so sustained work needs a higher plan.

Alternatives

Claude Code: Anthropic’s agent with similar surface coverage, deep customization and strong team controls, limited to Claude models.
Cline: an open-source, model-agnostic agent for VS Code, JetBrains and the terminal that uses your own keys.
Amp: a multi-model agent with cloud “orbs” that can run on your existing ChatGPT subscription.
GitHub Copilot: an assistant-first option if your team lives in GitHub and wants agent features inside an existing Copilot plan.

Final verdict

OpenAI Codex offers a capable, well-sandboxed agent with an open-source CLI, wide editor support and parallel cloud tasks, included in plans many developers already pay for. Budget time to track its fast-changing models and commands, and treat the published usage ranges as estimates, not guarantees. If those trade-offs are acceptable, it is one of the easiest coding agents to adopt.

Visit OpenAI Codex

FAQ

Is OpenAI Codex free?

Codex is included in every ChatGPT plan, including the Free and Go tiers, where it uses GPT-6 Luna and is meant for quick or lightweight coding tasks. Plans from Plus upward add GPT-6.1 Sol, OpenAI’s recommended model for complex coding, and larger allowances, although on Enterprise and Edu an admin must enable Sol first. Check OpenAI’s Codex pricing page for current limits, which change often.

Is my code used to train OpenAI’s models?

On individual plans it can be, unless you opt out. OpenAI’s help center says content from its services for individuals, such as ChatGPT and Codex, may be used to train its models, and that turning off “Improve the model for everyone” in ChatGPT’s Data controls, or using the Privacy Portal, covers Codex tasks; Codex also has a separate “Include environments” setting for training on full environments. By default, OpenAI says it does not use inputs or outputs from ChatGPT Business, Enterprise, Edu or the API to improve its models.

Does Codex work with VS Code and JetBrains?

Yes. The Codex IDE extension works in Visual Studio Code, VS Code Insiders, Cursor and Windsurf, while JetBrains IDEs (through AI Chat) and Xcode have their own integrations. Unlike the CLI, the IDE extension is not open source.

Can I use Codex with an API key instead of a ChatGPT plan?

Yes, with fewer features. Signing in with an OpenAI API key works for the CLI, SDK and IDE extension and is billed at OpenAI API token prices. Cloud features such as Codex Cloud are not available that way and require a ChatGPT sign-in.

Does Codex run on Windows?

Yes. OpenAI says Codex runs natively on Windows through the desktop app, the CLI and the IDE extension, with native sandbox implementations, while WSL2 uses the Linux sandbox. It also runs on macOS and Linux.

OpenAI Codex vs Claude Code: which should I choose?

Both cover the terminal, editors and cloud tasks, so the choice often comes down to models and subscriptions. Codex is built around OpenAI models and included in ChatGPT plans, while Claude Code is Anthropic’s agent, limited to Claude models, with deep customization and strong team controls. In our assessment, start with whichever matches the subscription your team already pays for, and consider a model-agnostic agent such as Cline if you want to switch providers freely.

OpenAI Codex Review (2026): CLI, IDE Extension and Cloud Agent