2026 Review

Codex

OpenAI's coding agent across CLI, IDE, cloud, and GitHub — included with paid ChatGPT plans.

CodingFreemiumAPIMainstream80% positiveAgent

Facts checked Aug 6, 2026

About Codex

Codex is OpenAI's agentic coding product: an open-source CLI, an IDE extension for VS Code and its forks, cloud agents that work tasks in parallel sandboxes, and automatic GitHub code review, all powered by the GPT-5 Codex model family. It's bundled with ChatGPT plans — limited use on Free, light quotas on Go at $8/month, full access on Plus at $20/month, and 5x/20x quotas on Pro at $100/$200/month — with usage metered in credits since April 2026.

At a glance

Founded: 2015
Headquarters: San Francisco, US
Platforms: CLI, IDE plugins, Web, iOS

Integrates with

GitHubVS CodeSlack

Main competitors

Claude Code Cursor GitHub Copilot

Best for

Parallel task delegationAutomated code reviewBug fixingFeature developmentLegacy refactors

Pros

Included in ChatGPT plans many developers already pay for — full access at $20/month
Cloud sandboxes make delegating several tasks at once genuinely practical
CLI is open source and scriptable
Free and $8 Go tiers let you try real agentic coding for almost nothing

Cons

Credit metering (introduced April 2026) makes heavy-use costs harder to predict
Agent runs are thorough but often slower than Claude Code's
Locked to the OpenAI ecosystem — no Anthropic or Google models

Our verdict

Best for: Developers on ChatGPT plans who want cloud-delegated coding tasks and automatic PR review without a second subscription.

Skip if: You want the strongest single-run agentic output regardless of ecosystem — Claude Code and Cursor rate higher with reviewers.

Our review

Codex's pitch is delegation at scale: instead of pairing with one agent in a terminal, you toss three or four tasks at cloud sandboxes from the web, your IDE, Slack, or your phone, then come back to review diffs. For teams living in GitHub, the automatic code review is quietly the stickiest feature — it catches real bugs on pull requests without anyone changing their workflow. And because it rides on ChatGPT plans, the marginal cost for millions of Plus subscribers is zero.

The knocks are pace and lock-in. Codex runs are careful and test-heavy but noticeably slower than Claude Code on comparable tasks, and the April 2026 move to credit-based metering — with a Pro split into $100 and $200 tiers — reintroduced the billing anxiety flat plans had solved. You're also all-in on OpenAI models; rivals let you pick Claude, GPT, or Gemini per task.

Verdict: the best value in agentic coding for anyone already inside the ChatGPT ecosystem, and the strongest cloud-delegation workflow of the major agents. Pick Claude Code if peak output quality and extensibility are worth a second subscription; pick Codex if one bill and parallel background work matter more.

What reviewers elsewhere say

80% positive — Praised as the best-value agentic coder on a $20 Plus plan and for careful, thorough diffs; criticized for slower runs and the 2026 shift to credit metering.

Sentiment estimated from reviews on Reddit, Product Hunt, G2, as of Aug 2026. Third-party scores are point-in-time snapshots. We don't host user reviews.

Key features

Cloud agents run multiple tasks in parallel, each in an isolated sandbox
Open-source CLI for terminal-driven agentic coding
IDE extension for VS Code, Cursor, and other VS Code forks
Automatic GitHub code review that flags bugs before merge
Start and monitor tasks from ChatGPT on iOS or from Slack
GPT-5 Codex models tuned for long-horizon engineering work

Codex as an agent

AgentRuns unattendedSomewhat predictable cost

Runs on: Your machine, Vendor cloud
Your own model key: Supported

What you give it access to

Your files
A shell on the machine it runs on
Code repositories

What to watch for

The cloud sandbox works from a snapshot of your repo, so results can assume stale context
Local runs execute commands against your real working tree
Included allowances differ sharply between ChatGPT tiers and are easy to exhaust

How it bills

Subscription + metered overage. One task here means one task handed to the CLI or a cloud sandbox → a diff you review before merging. ChatGPT plans include agent usage against a rolling limit; running it on an API key meters tokens per run.

Agent capabilities checked August 2026. Not hands-on tested — these are the vendor's documented capabilities, read carefully. Limits and permissions change between checks, so confirm approval settings and any spending cap before you connect an account. Compare every AI agent.

How to use Codex

Sign in with any ChatGPT account — Plus recommended for full access.
Install the Codex CLI or the IDE extension, or open Codex on the web.
Point it at a repo and describe a task; cloud tasks run in isolated sandboxes.
Queue several tasks in parallel and review the resulting diffs and PRs.
Enable GitHub code review so Codex checks pull requests automatically.
Watch credit usage in settings and step up a tier if you hit weekly limits.

Pricing

Free$0

ChatGPT Free includes limited Codex CLI usage.

Go$8/mo

Light Codex quotas on the budget ChatGPT plan.

Plus$20/mo

Full Codex on web, CLI, IDE extension, and iOS with standard credit quotas.

Pro 5x$100/mo

5x Plus usage; the Pro plan split into 5x and 20x tiers in April 2026.

Pro 20x$200/mo

20x Plus usage for parallel, all-day agent work.

Business$25/user/mo

$20/user/month billed annually. Workspace admin controls and shared access.

Pricing verified August 2026 · prices as listed by the vendor; check the official site for current offers.

Try Codex

Frequently asked questions

What is Codex?

Codex is OpenAI's agentic coding product. It spans an open-source CLI, an IDE extension, cloud agents that work on tasks in parallel sandboxes, GitHub code review, and iOS access — all driven by GPT-5 Codex models and bundled with ChatGPT plans.

How much does Codex cost?

Codex comes with ChatGPT plans: limited use on Free, light quotas on Go ($8/month), full access on Plus ($20/month), 5x or 20x quotas on Pro ($100 or $200/month), and Business at $25/user/month ($20 billed annually). Since April 2026, usage is metered in credits.

Can I use Codex for free?

Yes — ChatGPT Free includes limited Codex CLI usage, enough to evaluate it on small tasks. Sustained daily work realistically needs Plus or higher.

Does Codex have an API or SDK?

The Codex CLI is open source, OpenAI ships a Codex SDK for embedding the agent in custom workflows, and the underlying GPT-5 Codex models are available through the OpenAI API.

Codex vs Claude Code — which is better?

Codex wins on value if you already pay for ChatGPT, and on cloud parallelism plus built-in GitHub review. Claude Code generally rates higher on raw agentic output quality and customization. Ecosystem preference decides it for most teams.