Every score is worked out from the vendor's own pages · How we score →Disclosure
SAASINSPECTOR
OpenAI Codex logo

OpenAI Codex Review

OpenAI Codex is a coding agent that works from the terminal, your editor, a desktop app and the cloud. It edits files, runs commands in a sandbox and sends finished work back as pull requests. It is best for developers who already use ChatGPT and want an agent included in that plan.

Visit OpenAI Codex →From $8/mo

Research-based review. Features and prices are checked on the vendor's own website, and the score is worked out from those facts. We haven't tested it hands-on yet.

The verdict

Worth it if you already pay for ChatGPT and want a coding agent in the terminal or cloud; the catch is capped weekly usage on lower plans and GPT-6 models.

Pros

  • The Codex CLI is open source under Apache-2.0 and runs on macOS, Windows and Linux.
  • Cloud tasks on Plus and above end in a commit or a pull request, and @codex review works on pull requests.
  • Built-in Ollama and LM Studio providers let you run local models, and you can bring your own API key.
  • It is included in ChatGPT Free, and Go costs $8 a month, so you can start with no separate subscription.

Cons

  • Plus gives a few focused coding sessions each week, and Go is meant for lightweight coding tasks.
  • The IDE extension does not offer as-you-type completion.
  • Codex runs OpenAI's own GPT-6 Sol and Luna models, and the agent cannot run on your own infrastructure.
From $8/moFree plan Yes
7.2/10

Spec Score

OpenAI Codex against 13 AI coding assistants
#6
of 13 AI coding assistants
+0.1
vs the average
88%
from published facts
Ahead of other AI coding assistants
Where it works+2.7Pricing+1.5
Behind other AI coding assistants
Privacy and security-1.4Agents and your codebase-1.1
Every criterion
Tap a row for the facts
Agents and your codebase6.2
-1.1 vs 7.3 avg15.5 of 25 points
Agents that work in the cloud and open pull requestsyes3 / 3
Understands your whole codebasenot published, half points1.3 / 2.5
Reviews pull requestsyes2 / 2
Suggests code as you typeno0 / 2.5
Why these numbers

Agents that work in the cloud and open pull requests: Plus and above; tasks end in a commit or pull request.

Reviews pull requests: @codex review on pull requests and automatic reviews.

Suggests code as you type: The IDE extension opens a Codex chat sidebar with the open files as context; no as-you-type completion is described.

Pricing10.0
+1.5 vs 8.5 avg20 of 20 points
Monthly price, cheapest paid plan$8/mo6 / 6
Free planyes3 / 3
Free trialdoes not applyn/a
Why these numbers

Monthly price, cheapest paid plan: Plan: Go. Plus ($20) is the first plan with Codex on the web, in the CLI and in the IDE extension.

Free plan: Codex is included in ChatGPT Free, with GPT-6 Luna in the desktop app, subject to rollout.

Model freedom6.5
-0.2 vs 6.7 avg13 of 20 points
AI model families you can choosegpt1.5 / 5
Bring your own API keyyes3 / 3
Runs models on your own machineyes2 / 2
Why these numbers

AI model families you can choose: Custom model providers can use any Responses API compatible endpoint.

Runs models on your own machine: Built-in Ollama and LM Studio providers.

Privacy and security5.0
-1.4 vs 6.4 avg10 of 20 points
Does not train AI on your codeunless you opt out1.5 / 3
Option to keep no copy of your codenot published, half points1 / 2
Runs on your own servers or cloudno0 / 2
SOC 2yes2 / 2
Legal cover for generated codenot published, half points0.5 / 1
Why these numbers

Does not train AI on your code: Individual plans have a training setting; Business, Enterprise and API data is not used for training by default.

Runs on your own servers or cloud: Codex CLI runs on the user's machine and cloud tasks run in OpenAI-hosted environments; no customer-hosted option is documented for the agent.

SOC 2: Type 2, covers the API and ChatGPT business products.

Where it works9.0
+2.7 vs 6.3 avg13.5 of 15 points
Editors and terminals it works inclivscodejetbrainsxcode7 / 8
Open sourceyes2 / 2
Why these numbers

Editors and terminals it works in: Counted from platforms, plus its own editor where it is one.

Open source: The Codex CLI is open source (Apache-2.0). The IDE extension, cloud and desktop app are not.

yesnonot published average for AI coding assistants
How the Spec Score works

Scored from what OpenAI Codex publishes on its own site. Not a hands-on test.

Compared with 14 AI coding assistants. Facts checked 14 Oct 2026.

A fact the vendor does not publish gets half the points, or the typical value for a number, and says so. It never counts as a no. How we score.

OpenAI Codex is OpenAI's coding agent, available as an open-source command line tool, editor extensions, a desktop app and cloud tasks. It is not sold on its own: it comes with a ChatGPT plan, from Free up to Enterprise. What decides whether it suits you is how much coding you do each week, because the usage you get depends on the plan.

OpenAI Codex homepage screenshot
OpenAI Codex, Homepage

What is OpenAI Codex?

Codex is made by OpenAI, headquartered in San Francisco, United States. The command line tool runs on macOS, Windows and Linux. Editor extensions cover VS Code, Cursor, Windsurf, JetBrains and Xcode, and Codex also runs in a ChatGPT desktop app, on the web, in the ChatGPT iOS app, and through GitHub, GitLab and Slack.

The Codex CLI is open source under the Apache-2.0 licence. The IDE extension, the cloud service and the desktop app are not open source. Codex is not an editor of its own.

What Codex does in your code

Codex works as an agent. It edits files and runs commands in a sandbox, and an approval policy sets what needs your confirmation. The IDE extension opens a Codex chat sidebar with your open files as context, and it does not offer as-you-type completion.

Cloud tasks run in OpenAI-hosted environments and end in a commit or a pull request. They are on Plus and above. For review, you can write @codex review on a pull request, and automatic reviews are available too. The agent itself has no option to run on your own infrastructure.

Codex runs OpenAI's own GPT-6 Sol and Luna models. You can use your own API key, and custom model providers can point Codex at any endpoint compatible with the Responses API. Built-in Ollama and LM Studio providers let you run local models. Codex supports MCP, and the Codex SDK and the non-interactive codex exec command let you script it.

Beyond VS Code there are Xcode and JetBrains integrations, and the terminal CLI gives you the agent without an editor. Codex also has GitHub, GitLab and Slack integrations.

Working with other people

Business and Enterprise workspaces come with admin controls, analytics and audit exports. SAML single sign-on is on the Business plan and above, and Business is priced for 2 or more seats.

OpenAI Codex pricing: what each plan gives

OpenAI Codex pricing screenshot
OpenAI Codex, Pricing

Codex is included in ChatGPT plans, so the price is the price of the plan. Usage limits vary by plan, model and task size, and ChatGPT credits extend them. If you use an API key instead, you pay API token prices.

The Free plan lets you explore Codex on quick coding tasks with GPT-6 Luna at standard speed in the desktop app, subject to rollout, and it has no image generation. Go costs $8 a month and is meant for lightweight coding tasks. Plus costs $20 a month, gives a few focused coding sessions each week, and is the first plan with Codex on the web, in the CLI, in the IDE extension and on iOS.

Pro starts at $100 a month with higher usage limits, and a $200 tier also exists. Business costs $25 per user a month, or $20 per user a month billed annually, for 2 or more seats. Enterprise is priced by custom quote from sales.

Privacy and security

On individual plans, you decide whether your data is used for training and model improvement. Data from Business, Enterprise and the API is not used for training by default.

OpenAI has a SOC 2 Type 2 examination covering its API and ChatGPT business products, and it maintains ISO/IEC 27001:2022 and ISO/IEC 27701:2019 certifications. For GDPR, OpenAI can execute a Data Processing Addendum for ChatGPT Business, ChatGPT Enterprise and the API. A Business Associate Agreement is offered for qualifying business products, and the Codex docs include a HIPAA configuration for Work.

Codex support runs through GitHub issues and discussions, with a Help Center at help.openai.com.

Who OpenAI Codex suits

Codex suits developers who already pay for ChatGPT and want a coding agent in the terminal, the editor or the cloud without a separate subscription. Pro is there for heavier use, with higher usage limits. It also suits people who want an open-source CLI, a choice of local models through Ollama and LM Studio, or cloud tasks that finish as pull requests.

It suits less well heavy daily coders on the lower plans, since Plus gives a few focused sessions each week and Go is for lightweight tasks. It also suits less well anyone who wants Claude or other vendors' models, because Codex runs OpenAI's GPT-6 models, and anyone who wants as-you-type completion in the editor.

Frequently Asked Questions

How much does OpenAI Codex cost?

Codex is included in ChatGPT plans. Free costs nothing, Go is $8 a month, Plus is $20 a month, Pro starts at $100 a month, and Business is $25 per user a month or $20 billed annually. Enterprise is a custom quote.

Is there a free plan?

Yes. Codex is included in ChatGPT Free for quick coding tasks with GPT-6 Luna at standard speed in the desktop app, subject to rollout, with no image generation.

Which models does Codex run?

Codex runs OpenAI's GPT-6 Sol and Luna models. You can also use your own API key, point a custom provider at a Responses API compatible endpoint, or use the built-in Ollama and LM Studio providers.

Does OpenAI train on my code?

On individual plans there is a training setting, so you decide whether your data is used for training and model improvement. Business, Enterprise and API data is not used for training by default.

OpenAI Codex is featured in

Alternatives to OpenAI Codex

See all OpenAI Codex alternatives →

Other AI Developer Tool options we've reviewed.

GitHub Copilot logo

GitHub Copilot

Individual developers, teams, and enterprises seeking AI-powered code generation and assistance inside their existing IDEs

8.5
Rank #1 of 13 AI coding assistants
Best:Agents and your codebase10
Weakest:Privacy and security6.5
Free plan: yesFree trial: yesAPI: yes
FreeFree plan
100% from published facts · checked 14 Sep
→
JetBrains logo

JetBrains

Professional software developers using JetBrains IDEs seeking AI-powered code completion, agents, and enterprise governance

7.1
Rank #9 of 13 AI coding assistants
Best:Pricing10
Weakest:Where it works4
Free plan: yesFree trial: yesAPI: yes
From $10Free plan
96% from published facts · checked 14 Sep
→
Cursor logo

Cursor

Software developers and engineering teams building complex, large-scale software with AI-assisted and agentic coding workflows

8.4
Rank #2 of 13 AI coding assistants
Best:Agents and your codebase10
Weakest:Where it works7
Free plan: yesFree trial: noAPI: yes
From $20Free plan
100% from published facts · checked 14 Sep
→
Tabnine logo

Tabnine

Enterprise development teams and individual developers who prioritize privacy, security, and compliance

7.5
Rank #4 of 13 AI coding assistants
Best:Privacy and security10
Weakest:Pricing3
Free plan: noFree trial: noAPI: yes
From $39No free plan
100% from published facts · checked 14 Sep
→
Claude logo

Claude

Writers, researchers and developers who want voice, memory, projects and code execution free, with deep research on paid plans

7.8
Rank #8 of 10 AI assistants
Best:Where it works10
Weakest:Pricing6
Free plan: yesFree trial: noAPI: yes
From $20Free plan
100% from published facts · checked 21 Sep
→
CodeRabbit logo

CodeRabbit

Engineering teams and developers seeking automated AI-powered code review on GitHub and GitLab

9.1
Rank #1 of 5 code review and security tools
Best:Review quality10
Weakest:Privacy and deployment7
ISO 27001: noHIPAA: no
From $24Free plan
100% from published facts · checked 14 Sep
→

User reviews

Review OpenAI Codex

Your rating

Reviews are moderated and appear once approved.