Disclosure: some links on this page may be affiliate links; we may earn a commission at no extra cost to you. Read the disclosure.

By the toolist.ai editors. Researched from Codex's own pages and hands-on tests published by others, not hands-on tested by toolist.ai. Updated October 13, 2026.

Codex review: pricing, features and what testers found

Codex is OpenAI's coding agent: it reads, edits and runs code from the ChatGPT desktop app, a terminal CLI, an IDE extension and cloud tasks that can open pull requests, and it reviews pull requests on GitHub. It comes with every ChatGPT plan, including Free; paid plans start at $8 a month (Go), and Plus at $20 is the cheapest plan OpenAI lists with the CLI, IDE extension, cloud tasks and GPT-6.1 Sol.

Our take

Codex suits developers who already pay for ChatGPT and want to hand well-scoped coding tasks to an agent in the desktop app, terminal, editor or cloud: Plus at $20 a month is the cheapest plan OpenAI lists with the CLI, IDE extension, cloud tasks and GPT-6.1 Sol. Testing it on a real Python project, The New Stack's Jessica Wachtel saw it fix a bug from a GitHub issue in 3 minutes and give a useful pull request review, and Cybernews' Adam Koscielak called it the most user-friendly coding assistant he had used, though his more complex game still did not play properly after several rounds of fixes. Zapier's Miguel Rebelo recommends Codex over Cursor for delegating whole tasks and Cursor for staying close to the code. The catch is that OpenAI publishes no fixed allowance, only an estimated 15 to 160 local GPT-6.1 Sol messages per five hours on Plus, weekly limits may also apply, and extra usage is bought as credits with no published dollar price. On Free, Go, Plus and Pro your Codex content may train OpenAI's models unless you opt out; Business seats ($25 a user a month, or $20 billed yearly) are not used for training by default.

Based on Codex's published plans and the hands-on tests linked below, not on a test of our own.

Good for

  • ChatGPT subscribers who want a coding agent with no extra subscription
  • Handing well-scoped bug fixes and maintenance to cloud tasks and reviewing the result as a pull request
  • Automatic pull request reviews on GitHub
  • Scripted and CI runs paid per token with an API key

Watch out for

  • OpenAI publishes no fixed allowance. Plus and Standard Business seats get an estimated 15 to 160 local messages per five hours on GPT-6.1 Sol (5 to 45 on GPT-6 Astra, 350 to 3,000 on GPT-6 Luna), cloud tasks may use more, and weekly limits may also apply; Pro plans currently have no five-hour limit, but OpenAI does not say how much more each Pro tier includes.
  • Fast mode uses included usage 2.5 times as fast as Standard and Ultrafast 8 times, and image generation uses it 3 to 5 times faster on average; Codex shares its allowance and credits with ChatGPT Work, Excel, PowerPoint and Word.
  • Past your limits, Plus and Pro users (and a limited group of Free and Go users) can buy credits, but OpenAI's public pages give no dollar price per credit; credits expire 12 months after purchase and are non-refundable unless the law requires it or for billing errors, unauthorized use or confirmed service failures. For credit-billed Business and Enterprise workspaces, OpenAI's rate card estimates Codex at about $100 to $200 per developer a month on average.
  • On Free, Go, Plus and Pro, OpenAI may use your Codex content, including screenshots taken by computer use, to train its models unless you turn training off in ChatGPT data controls; Business, Enterprise, Edu and API data is not used for training by default.
  • Payments are generally non-refundable: an unused individual subscription may get a discretionary refund if you ask within seven days of the charge, EU, UK and Turkey residents get a prorated refund within 14 days, and unused Business seats are not refunded. Pro is billed monthly only.
  • On Free and Go, the Codex pricing page lists only GPT-6 Luna at Standard speed in the desktop app, subject to rollout (ChatGPT's pricing page adds limited GPT-5.6 Terra in Work and Codex on desktop), and Codex Cloud is not included. Allowances also change: Pro 200's allowance was cut for new subscriptions in late September 2026, and earlier subscribers move to the lower allowance after October 29, at the same $200 price.

What Codex does

Price

Free plan: ChatGPT Free includes limited Codex use: GPT-6 Luna at Standard speed in the ChatGPT desktop app, subject to rollout, with no cloud tasks or image generation; OpenAI publishes no fixed allowance.

PlanBilled monthlyBilled yearly, per monthIncludes
Free free not listed Limited Codex use: GPT-6 Luna at Standard speed in the ChatGPT desktop app, subject to rollout; no cloud tasks or image generation
Go $8 not listed Codex for lightweight tasks: GPT-6 Luna at Standard speed in the desktop app, subject to rollout; no cloud tasks; billed monthly
Plus $20 not listed Codex in the desktop app, on the web, in the CLI, IDE extension and on iOS; cloud tasks, GitHub code review and Slack; GPT-6.1 Sol, GPT-6 Astra, Sol and Luna; an estimated 15 to 160 local GPT-6.1 Sol messages per five hours, weekly limits may apply; extra usage with credits
Pro 100 $100 not listed Everything in Plus with more usage and currently no five-hour limit (OpenAI does not say how much more); Computer History; billed monthly only
Pro 200 $200 not listed More usage than Pro 100; the allowance was lowered for new subscriptions in late September 2026, and earlier subscribers keep the old one only until October 29, 2026; billed monthly only
Pro 500 $500 not listed The highest included usage of the Pro plans, and Ultrafast mode for GPT-6 Astra and GPT-6.1 Sol, which uses included usage at 8 times the Standard rate; billed monthly only
Business (Standard seat) $25 $20 Per user, at least 2 seats and up to 200: ChatGPT and Codex with the same estimated limits as Plus, larger cloud machines, SAML SSO and MFA, workspace credits for extra usage; business data not used for training by default
Business (Premium seat) $125 $100 Per user: five times a Standard seat's usage and no five-hour limit; can be mixed with Standard seats
Enterprise & Edu not listed not listed Contact sales: priority processing, SCIM, EKM, RBAC, audit logs through the Compliance API, data retention and residency controls; with flexible pricing, a pooled credit balance and no fixed rate limits, otherwise per-seat limits like Plus
OpenAI API key not listed not listed Pay per million tokens at API rates: GPT-6.1 Sol $2 input and $10 output ($0.10 cached input), GPT-6 Luna $0.10 and $0.50, GPT-6 Astra $10 and $50, gpt-5.3-codex $1.75 and $14; prompts over 272,000 input tokens cost more. Codex in the CLI, SDK or IDE extension, with no cloud features such as GitHub code review or Slack

Prices last checked: 2026-10-16 on Codex's pricing page. Full breakdown: Codex pricing.

Price history

No price change since we first recorded these prices on October 13, 2026. We re-check them every week and list each change here.

What hands-on testers found

We have not tested Codex ourselves. These are the published tests we found that used it hands-on, newest first; follow the links for the full reviews.

Zapier · Miguel Rebelo · May 13, 2026

What they tested: Codex and Cursor side by side over dozens of hours of personal use and testing, comparing the coding workflow, codebase context, parallel agents, models, token use and pricing.

  • Codex is built for delegation: you brief a task and review the result, where Cursor keeps you editing in an IDE alongside the AI.
  • Each cloud task starts from a fresh clone that Codex must search before it can answer, which isolates work cleanly but starts slower than Cursor's standing codebase index; by default the network is cut off once dependencies are installed.
  • He recommends Codex for handing off whole tasks and reviewing pull requests, and Cursor for staying close to the code; at the time the web and desktop apps ran only OpenAI models.

The New Stack · Jessica Wachtel · May 7, 2026

What they tested: A day with the Codex desktop app on the free plan, fixing a bug in the open-source HTTPie Python CLI with the in-app browser, computer use and pull request review.

  • Pointed at the GitHub issue in the in-app browser, Codex fixed the bug in 3 minutes: it traced it to three files, added a regression test, ran the tests and left unrelated changes alone.
  • Computer use refused to drive Terminal for security reasons and fell back to its own shell, and a Finder screenshot failed.
  • Its pull request review confirmed the fix, cited library docs and flagged a gap in test coverage, though the sandbox blocked the full test suite; she called it the most complete Claude Code alternative.

Cybernews · Adam Koscielak · March 2, 2026

What they tested: The Codex macOS app building a time-management app and a curling game, finding bugs planted by another model, and rewriting code into other languages.

  • The calendar, kanban and wiki app worked with only very minor bugs, and he called Codex the most user-friendly coding assistant he had used.
  • The curling game still played wrong after several rounds of fixes, and it sometimes misread layout requests such as a responsive design.
  • It found the planted bugs, but he found cloud mode through GitHub unintuitive, and the desktop app was macOS-only at the time.

Their score: 4.5 out of 5

What users say

Test result

toolist.ai has not run its own test of Codex; the findings above come from the published tests linked. See how we research and test.

Best coding tools · Codex pricing

Do you make Codex? Claim this listing to correct it; it's free. To see whether ChatGPT, Claude, Gemini and Perplexity recommend it, and how many readers we send you: toolist.ai for AI companies.

Sources