Claude
Anthropic's AI assistant built for long-context analysis, coding, and safe reasoning
FREEMIUM Verified JUL 3, 2026
OpenAI Codex
OpenAI's software-engineering agent — open-source CLI, IDE extension, and cloud tasks, bundled with every ChatGPT plan
OpenAI Codex is OpenAI's software-engineering agent, spanning an open-source Apache-2.0 CLI, an IDE extension, GitHub code review, and cloud tasks that run in parallel sandboxes. It is bundled with every ChatGPT plan: limited access on Free, Plus at $20/month for everyday use, and Pro from $100/month with 5x or 20x rate limits. Heavy users reportedly land at $100–200/month in real usage. Best for developers already paying for ChatGPT; for frontier coding work many teams still reach for Claude first.
Verified JUL 6, 2026 FREEMIUM Live
What is OpenAI Codex?
OpenAI Codex is OpenAI’s software-engineering agent — “one agent for everywhere you code,” in OpenAI’s framing. It shows up in four surfaces that share a single sign-in: an open-source CLI (Apache-2.0, written in Rust, roughly 96k GitHub stars) that runs in your terminal on macOS, Linux, and Windows; an IDE extension; cloud tasks that execute in sandboxed environments preloaded with your repository, several in parallel; and an automated code reviewer for GitHub pull requests. A Codex SDK lets teams script the same agent inside CI pipelines. Under the hood it runs OpenAI’s GPT-5.5-family models — GPT-5.5, GPT-5.4, and GPT-5.4 mini — and only those; there is no option to route work to other providers.
The distribution model is the unusual part: Codex is bundled with every ChatGPT plan rather than sold separately. Free accounts get limited access, Plus ($20/month) covers everyday use with 15–80 GPT-5.5 local messages per 5-hour rolling window, and Pro starts at $100/month with a choice of 5x or 20x the Plus limits. In April 2026 OpenAI moved Codex metering from per-message to API-token-aligned accounting, which made light sessions cheaper but heavy agentic sessions harder to predict — third-party analyses put realistic heavy-use spend at $100–200 per developer per month. OpenAI reported around 4 million developers active on Codex weekly at the GPT-5.5 launch.
Who is it for?
Codex makes the most sense if you are already inside the OpenAI ecosystem: the marginal cost of trying it on an existing Plus subscription is zero, and the CLI is a free install. It is less compelling if you want to choose your model per task — multi-model editors like Cursor let you switch between providers, and for frontier agentic coding many developers pair Codex with or prefer Claude and Claude Code outright.
- Existing ChatGPT Plus and Pro subscribers who write code and want an agent without adding another subscription — Codex is already included in what they pay.
- Terminal-first developers who want an open-source, auditable agent they can install with npm or Homebrew and run against either a subscription or a plain API key.
- Teams drowning in pull requests that want automatic first-pass code review on GitHub before human reviewers spend time on a diff.
- Developers who batch background work — dispatching parallel cloud tasks for bug fixes, test coverage, and refactors while continuing to work locally.
- Platform engineers embedding an agent into CI via the SDK on usage-based API pricing rather than seat licenses.
How much does OpenAI Codex cost?
Starting price: $8/mo · Free tier: yes · Model: freemium
Pricing verified JUL 6, 2026
Price history tracked from June 2026
| Plan | Price | Includes |
|---|---|---|
| Free | $0 | Limited Codex access included with a free ChatGPT account · Open-source CLI installable without any subscription · Enough for sampling and light personal use (reported) |
| Go | $8/mo | Lightweight Codex usage on the cheapest paid ChatGPT plan · Same surfaces (CLI, IDE, cloud) with lower rate limits |
| Plus | $20/mo | 15–80 GPT-5.5 local messages per 5-hour rolling window · 10–60 cloud tasks and 20–50 code reviews per 5-hour window (reported) · Extra credits purchasable when limits run out |
| Pro 5x | $100/mo | 5x Plus limits: 75–400 GPT-5.5 local messages per 5-hour window · Launched April 2026 as the mid-step between Plus and Pro 20x |
| Pro 20x | $200/mo | 20x Plus limits: 300–1,600 GPT-5.5 local messages per 5-hour window · 200–1,200 cloud tasks per window (reported) · For full-time agentic development |
| Business | $25/user/mo | Reported at $25/user/month ($20 annual); some sources describe pay-as-you-go seat pricing · Workspace admin controls |
| Enterprise / Edu | Custom | Custom pricing and pooled usage · Access tokens for workspace management |
What are OpenAI Codex's key features?
- Cloud tasks: delegate coding jobs to sandboxed environments preloaded with your repo, multiple tasks in parallel
- Codex CLI: open-source (Apache-2.0) terminal agent written in Rust, installed via npm, Homebrew, or a shell script
- IDE extension that brings the same agent into your editor
- GitHub code review that flags bugs on pull requests automatically
- Codex SDK for scripting the agent inside CI and custom workflows
- Runs GPT-5.5-family models (GPT-5.5, GPT-5.4, GPT-5.4 mini) with per-model rate limits
- Sign in with your ChatGPT account — no separate billing on subscription plans — or bring an API key for pay-per-token usage
What people use OpenAI Codex for
- 01 Delegating background work — bug fixes, refactors, test writing — to parallel cloud sandboxes while you keep coding locally
- 02 Terminal-first agentic coding with the open-source Rust CLI on macOS, Linux, or Windows
- 03 Automated code review on GitHub pull requests before a human ever looks at the diff
- 04 Embedding the agent into CI and automation pipelines via the Codex SDK
- 05 Codebase Q&A and onboarding — asking how unfamiliar code works before editing it
Pros and cons
| Pros | Cons |
|---|---|
| Bundled with every ChatGPT plan from Free to Enterprise — if you already pay $20/month for Plus, trying Codex costs nothing extra | Usage limits are opaque wide ranges (Plus: 15–80 GPT-5.5 messages per 5-hour window) that vary with task complexity, so it is hard to predict when you will hit a wall |
| CLI is genuinely open source (Apache-2.0, ~96k GitHub stars) and works standalone with an API key | The April 2026 switch to token-based metering made costs less predictable; heavy users reportedly spend $100–200/month in practice |
| One agent across terminal, IDE, cloud, and GitHub with a single sign-in | Locked to OpenAI models only — no option to route tasks to Claude or Gemini, unlike multi-model editors such as Cursor |
| Cloud tasks run in parallel, so you can dispatch several fixes at once instead of babysitting one session | Cloud tasks require granting OpenAI's sandboxes access to your repositories, a compliance hurdle for some organizations |
What are the best OpenAI Codex alternatives?
Cursor
AI-native code editor that runs autonomous agents to build, test, and review code
FREEMIUM Verified JUL 3, 2026
GitHub Copilot
AI coding assistant with inline completions, chat, and autonomous agent mode across major IDEs
FREEMIUM Verified JUL 3, 2026
Jules
Google's asynchronous AI coding agent that clones your repo, plans changes, runs in a cloud VM, and opens a reviewable pull request
FREEMIUM Verified JUN 30, 2026
Aider
Open-source AI pair programming in your terminal — bring your own LLM key and edit your git repo
OPEN-SOURCE Verified JUN 22, 2026
Devin
Autonomous AI software engineer from Cognition that takes whole engineering tasks from ticket to pull request
USAGE-BASED Verified JUN 20, 2026
See all OpenAI Codex alternatives →
Head-to-head comparisons
How people make money with OpenAI Codex
- Backlog-clearing freelance service — fire off parallel Codex cloud tasks on a client's bug backlog and price per closed ticket; a Plus plan's reported 10–60 cloud tasks per 5-hour window covers a day's batch
- Codex code-review setup for small teams as a fixed-fee consulting offer — the GitHub PR review integration runs on the ChatGPT plan the team already pays for, so the engagement is pure configuration labor
Frequently asked questions
Is OpenAI Codex free?
Partly. Codex is included with every ChatGPT plan, including the free tier, with limited usage. The open-source CLI can also be installed for free and used with an API key on pay-per-token billing. Sustained daily use realistically requires Plus at $20/month or higher.
Which ChatGPT plans include Codex?
All of them: Free ($0), Go ($8/month), Plus ($20/month), Pro from $100/month with a choice of 5x or 20x rate limits, plus Business, Edu, and Enterprise plans. Limits scale with the plan — Plus gets 15–80 GPT-5.5 local messages per 5-hour window, Pro 20x gets 300–1,600.
What models does Codex use?
Codex runs OpenAI's GPT-5.5 family: GPT-5.5, GPT-5.4, and GPT-5.4 mini, with additional variants like GPT-5.3-Codex on the rate card. Rate limits differ per model — the mini model allows roughly 4x more messages per window than GPT-5.5. It cannot use non-OpenAI models.
Is Codex CLI open source?
Yes. The CLI is Apache-2.0 licensed, written in Rust, and hosted on GitHub with roughly 96k stars. It installs via npm, Homebrew, or a one-line shell script on macOS, Linux, and Windows, and authenticates with either a ChatGPT account or an API key.
How is Codex different from GitHub Copilot?
Copilot lives primarily inside your editor as completions plus an agent mode; Codex is a standalone agent that also runs cloud tasks in parallel sandboxes and reviews GitHub pull requests. Codex is billed through ChatGPT plans starting at $20/month for meaningful limits, while Copilot has its own per-seat pricing.
Can I use Codex without a ChatGPT subscription?
Yes, two ways: the free ChatGPT tier includes limited Codex access, and the open-source CLI accepts an OpenAI API key with usage-based, pay-per-token pricing instead of plan limits. API-key mode is the route for CI pipelines and automation via the SDK.
What happens when I hit Codex usage limits?
Limits reset on a 5-hour rolling window (with weekly caps on some plans, per third-party reports). Plus and Pro users can buy extra credits to keep working, upgrade to Pro 5x ($100/month) or Pro 20x ($200/month), or switch the CLI to an API key and pay per token.