Codex vs GitHub Copilot: Where Each Agent Lives

If you last compared these two a few years ago, forget the old framing. Codex vs GitHub Copilot is no longer "an agent vs an autocomplete." Both now run in your editor, in your terminal, and in the cloud, and both open pull requests on their own, though which Codex surfaces you get depends on your ChatGPT plan.

What still separates them is where each one lives, which models it runs, and how you pay for it.

• Codex is OpenAI's coding agent. It comes with your ChatGPT account and runs as a CLI, an IDE extension, a desktop app, and cloud tasks, with surface access depending on your plan.
• GitHub Copilot is GitHub's AI platform. It has as-you-type completions, chat, an agent in your IDE, Copilot CLI, a cloud agent that works from issues, and pull request reviews.
• Model choice is the biggest practical gap. Copilot offers models from several vendors. Codex is built around OpenAI's models.
• You can use Codex inside Copilot. GitHub offers an OpenAI Codex agent in public preview, paid for through your Copilot plan.

Quick comparison

Why "autocomplete vs agent" is out of date

The confusion has a history. When GitHub launched Copilot in 2021, it was "powered by OpenAI Codex," a code model OpenAI built for completions. So for a while, "Codex vs Copilot" meant "the model vs the product built on it."

That model is gone. OpenAI's deprecations page lists the old code-davinci and code-cushman Codex models as shut down on March 23, 2023.

Today's Codex is a different thing with the same name: an agent that reads your repo, edits files, runs commands, and works on tasks in the cloud. Copilot grew the other way. It started as completions and added an agent mode, a CLI agent, and a cloud agent.

Here's how the two line up today:

So both are agents now. The useful question is which one fits how your team already works.

The real difference: where each one lives

Copilot lives in GitHub

Copilot's strength is that the work never leaves GitHub. You assign an issue to Copilot, or start a task from the agents panel or an @copilot comment on a pull request. The Copilot cloud agent researches the repo, plans, and makes changes in an ephemeral cloud environment where it can run tests and linters. Then it hands you a pull request.

Review stays in the same place. Copilot code review can be requested on any PR, and repo or org owners can turn on automatic reviews for everyone. Org admins control all of it through Copilot policies.

The trade-off: GitHub's docs say the cloud agent only works with repositories hosted on GitHub. If some of your code is on GitLab or Bitbucket, that part is out.

Codex lives in your ChatGPT account

Codex starts from ChatGPT, not from your code host. You sign in with ChatGPT, and on plans that include them the same account works across the CLI, the IDE extension, the desktop app, and the web.

Not every plan gets every surface. The Codex pricing page lists the Free and Go plans with GPT-6 Luna in the desktop app, subject to rollout, while the CLI, IDE extension, web, and cloud integrations belong to applicable paid plans.

The Codex CLI is open source under Apache 2.0. It runs locally, reads and edits your files, and runs commands. The IDE extension works in VS Code and compatible editors like Cursor and Windsurf. Codex's docs note that Xcode and JetBrains IDEs have their own integrations.

Codex cloud tasks, on plans that include cloud features, run in their own isolated workspace, so they keep going when your laptop is shut. You connect GitHub, pick repositories, review the result, and open a PR when you're ready. Codex also reviews pull requests on GitHub. You mention @codex review, or you turn on automatic reviews through the ChatGPT Codex connector.

So Codex touches GitHub too. The difference is that GitHub is one place it connects to, not the place it lives.

If you want a deeper tour of Codex's modes, sign-in, and AGENTS.md, see our OpenAI Codex guide.

Model choice: many vendors vs OpenAI-first

This is the clearest difference, and it doesn't change from month to month the way model names do.

Copilot is multi-vendor. GitHub's supported models list currently includes models from OpenAI, Anthropic, Google, Microsoft, Moonshot AI, and xAI. You switch models in the chat picker, and Copilot CLI has a /model command. Your plan decides which models you can use.

Codex is OpenAI-first. The Codex models page recommends OpenAI's current models for different kinds of work. You can point Codex at a custom provider or gateway, but it has to expose a compatible Responses API endpoint. That's possible, but it's a setup job, not a dropdown.

Note that Copilot's list includes OpenAI's Codex-tuned models too. If you want OpenAI models but live in GitHub, you don't need Codex for that. Model names on both sides change often, so check each vendor's current list rather than trusting a blog post, this one included.

Using Codex inside GitHub Copilot

This is the part most comparisons miss. Since GitHub announced Agent HQ, Copilot can run coding agents from other vendors next to its own.

Here's what GitHub's third-party agents docs say today:

• Two third-party agents are supported: Anthropic Claude and OpenAI Codex.
• They are in public preview.
• They're available on Copilot Pro, Pro+, Max, Business, and Enterprise, but have to be enabled in policy settings first.
• You can use them from the Agents tab, issues, pull request comments, GitHub Mobile, and VS Code.
• They consume GitHub Actions minutes and AI credits.

The OpenAI Codex agent page adds that the agent is built on the Codex SDK, runs on your existing Copilot subscription, and offers a short list of OpenAI models rather than the full Codex lineup.

So "use Codex in GitHub Copilot" is a real option, with limits. You get a Codex-powered agent inside GitHub's workflow, billed by GitHub. You don't get your ChatGPT account, your Codex CLI setup, or Codex's own cloud tasks. It's the same agent family in GitHub's house.

Pricing: how each one bills

Both vendors change prices often, so here are the billing models. For numbers, go to the linked pages.

Copilot uses per-seat plans with a monthly allowance of AI credits. Chat, agent, and review usage is charged by the model you use and the tokens it processes. Completions and next edit suggestions aren't billed in AI credits and stay unlimited on paid plans. Unused credits don't roll over. There's also a no-cost Copilot tier with a monthly cap on completions.

Codex comes with ChatGPT. Per the Codex pricing page, every ChatGPT plan from Free through Enterprise includes some Codex access, but not the same access. Free and Go get GPT-6 Luna in the desktop app, subject to rollout. The CLI, IDE extension, web, and cloud features belong to applicable paid plans. How much you can do depends on the model, the size of your tasks, and your plan. When you hit your limit, Plus and Pro can buy extra credits, and so can Business, Edu, and Enterprise workspaces on flexible pricing. The page doesn't say Free or Go can. You can also run Codex on an API key and pay standard API rates, which are billed separately and don't include cloud features such as GitHub code review or Slack.

Is Codex better than Copilot?

Neither wins outright. Here's where each is stronger.

Pick Codex if:

• You already have a ChatGPT plan that includes the surfaces you need (terminal, editor, desktop, cloud), and you want one account across them.
• You want an open-source CLI you can read, fork, and run in scripts or CI.
• You want an agent that isn't tied to one code host.
• You're happy with OpenAI's models.

Pick Copilot if:

• Your team lives in GitHub issues and pull requests, and you want agents, reviews, and policies in the same place.
• You want as-you-type completions. Codex doesn't sell itself as an autocomplete tool.
• You want to switch between models from several vendors.
• You need org-level admin controls over who uses which models and agents.

Use both if your team is on Copilot but some people prefer Codex. Enable the Codex agent in Copilot, or let those people run the Codex CLI with their ChatGPT accounts.

If Claude is also on your list, Claude Code vs GitHub Copilot covers that matchup. If you'd rather work in an AI-first editor, see Cursor vs Codex.

When the setup is the real problem

Codex and Copilot both assume you already have a repo that runs, a working terminal, and the context to judge a diff. For developers, that's fine.

But plenty of people get stuck before any of that. You saw a Codex workflow or an agent setup on X, it did exactly what you needed, and the README assumed three things you didn't have.

Taku is built for that gap. It's an AI-native desktop workspace where you mirror an AI app, skill, or workflow someone already got working, run it, and keep it. It isn't a coding agent and doesn't replace either tool. Taku is in Beta, and the Mac app is available now.

FAQ

Is Codex the same thing that powered the original GitHub Copilot?

No. Copilot launched in 2021 on a code model called OpenAI Codex. OpenAI shut those models down on March 23, 2023. Today's Codex is a separate coding agent that shares the name.

Can I use Codex in GitHub Copilot?

Yes, in public preview. GitHub offers an OpenAI Codex agent on Copilot Pro, Pro+, Max, Business, and Enterprise, once it's enabled in policy settings. It's billed through Copilot, using GitHub Actions minutes and AI credits, not your ChatGPT plan.

Does Codex do autocomplete like Copilot?

Codex is built around tasks and chat, not as-you-type suggestions. Its IDE extension is a panel you talk to. If inline completions matter to you, Copilot is the one built for them.

Is Codex CLI open source?

Yes. The Codex CLI repository on GitHub is licensed under Apache 2.0, so you can read the code, fork it, and run it in scripts.

Codex vs Copilot pricing: which is cheaper?

It depends on what you already pay for. Codex comes with ChatGPT plans, but surfaces and extra credits vary by plan: Free and Go are limited to the desktop app (subject to rollout), and credits are offered to Plus, Pro, and flexible-pricing Business, Edu, and Enterprise. An API key is billed separately and has no cloud features. Copilot is a per-seat plan with monthly AI credits, and completions aren't metered. Check both vendors' pricing pages before you decide.

Can Copilot use OpenAI models without Codex?

Yes. Copilot's model picker includes OpenAI models, including Codex-tuned ones, next to models from Anthropic, Google, and others.

What other tools should I compare?

GitHub Copilot alternatives covers the wider field. For Codex specifically, see the Codex guide linked above.