Head to head
Codex vs Copilot: OpenAI Codex vs GitHub Copilot Pricing, Credits and Agent Limits Compared
The sticker prices are a dollar apart. What separates these two is that one bills you a budget and the other bills you a rate, which decides whether you get an unexpected invoice or an unexpected wait. Every figure here was read from OpenAI's and GitHub's own documentation in August 2026.
No card to start · Flat $29 a month, no usage meter
Pick a task
Plan
- planning
Files changed
Test run
Pull request
You review and merge. Agentcode never merges on its own.
The short answer
Codex and GitHub Copilot are billed in different kinds of unit, and that is the decision. Copilot meters actions in AI credits with a published price of $0.01 each, so a Copilot month can be forecast and can also be overspent. Codex meters messages against a rolling five-hour window, so a Codex month cannot be overspent and cannot be priced either, because OpenAI publishes no dollar value for a ChatGPT credit. On list price they are almost identical: Copilot Pro $10, Pro+ $39, Max $100, Business $19 a seat; Codex Go $8, Plus $20, Pro from $100, Business $20 a user annually or $25 monthly. Copilot is the only one of the two with inline autocomplete, unmetered on every paid plan. Codex is the only one that reads GitLab and the only one you can run on a plain API key. And it is rarely either/or: Copilot runs OpenAI's Codex models, and GitHub bills the Codex VS Code extension against your Copilot allowance.
The difference every other comparison skips
Read the first two rows and you have the answer. Every "which is cheaper" verdict in this search result assumes the two prices measure the same thing. They do not. Copilot sells you a monthly budget of actions at a published cent apiece. Codex sells you a refilling rate of messages with no published dollar value at all.
Swipe to see more
| The question | OpenAI Codex | GitHub Copilot | Why it matters |
|---|---|---|---|
| What are you actually billed for? | Messages, against a rolling five-hour window | Actions, against a monthly credit allowance | This is the whole comparison in one row. Codex gives you a rate: so many messages every five hours, refilling continuously. Copilot gives you a budget: so many credits a month, spent per action, gone when they are gone. A rate cannot be overspent. A budget cannot make you wait. |
| Is there a published dollar price for one unit? | No | Yes, $0.01 per AI credit | OpenAI publishes credits per million tokens and says GPT-5.6 usage "averages 5-30 credits per message", but no dollar value for a ChatGPT credit. GitHub publishes a flat cent. So you can forecast a Copilot overage bill from the vendor docs and you cannot forecast a Codex one. |
| What happens when you hit the limit? | You wait, switch to a smaller model, or buy credits | You keep working and the bill grows | OpenAI lets you finish the turn you are on, then you wait out the window or buy more. GitHub charges $0.01 a credit past the allowance, so nothing stops, which is either the feature or the risk depending on who signs off your invoices. An admin can cap it with a budget. |
| Cheapest paid entry for one developer | $8 a month (Go) | $10 a month (Pro) | Close enough to be a tie, and both have a free tier above it. OpenAI positions Go for "lightweight coding tasks", so the honest like-for-like starting comparison is Codex Plus at $20 against Copilot Pro at $10 or Pro+ at $39. |
| Team seat, list price | $20 per user annually, $25 monthly | $19 per seat (Business), $39 (Enterprise) | A dollar apart on the annual commitment, and a quarter more on Codex if you refuse to commit for a year. OpenAI requires 2 or more users on Business. The list prices are so close that they should not decide this, which is exactly why the metering row above should. |
| Inline autocomplete as you type | Not offered | Unlimited and never metered on paid plans | The single cleanest functional split. GitHub states that "code completions and next edit suggestions are not billed in AI credits and remain unlimited for all paid plans". Codex has no autocomplete product at all. If ghost-text completion is what you want, only one of these two sells it. |
| Can you run it without a subscription? | Yes, on an API key, paying per token | No | Codex CLI, the SDK and the IDE extension all run on a plain API key, which OpenAI recommends for "automation in shared environments like CI". You lose the cloud features: no GitHub code review, no Slack. Copilot has no equivalent seatless mode. |
One consequence is worth stating on its own, because it decides who should sign off the purchase. A Copilot overrun is a finance problem you can cap with a budget. A Codex overrun is an engineering problem you cannot pay your way out of in the moment: you wait for the window, or you drop to a smaller model. If the credit half is new to you, we unpack it on what GitHub AI credits are and on why AI coding tools use credits at all.
Codex vs GitHub Copilot pricing, tier by tier
Lined up by who each tier is for rather than by price, because the two ladders do not have matching rungs. Copilot has no $8 tier and Codex has no $39 individual tier. Every price below is the vendor's published US list price.
Free
OpenAI Codex
$0
Explore Codex on quick coding tasks.
GitHub Copilot
$0
An allowance of AI credits, auto model selection only, and completions capped at 2,000 a month. Only for developers with no org licence.
Entry paid
OpenAI Codex
$8 (Go)
OpenAI scopes it to "lightweight coding tasks". There is no Copilot tier at this price.
GitHub Copilot
$10 (Pro)
1,000 base credits plus a 500 flex allotment, 1,500 total, and a selection of models.
Working developer
OpenAI Codex
$20 (Plus)
Codex on the web, CLI, IDE extension and iOS, cloud integrations including automatic code review and Slack, and the GPT-5.6 family. 25 to 200 Terra messages per five-hour window.
GitHub Copilot
$39 (Pro+)
3,900 base plus 3,100 flex, 7,000 total, and access to premium models. Also the only tier that unlocks GPT-5.4 nano in the Codex VS Code extension.
Heavy user
OpenAI Codex
$100 or $200 (Pro)
5x Plus limits at $100, 20x at $200. Adds GPT-5.3-Codex-Spark in research preview, on a separate usage limit that "may adjust based on demand".
GitHub Copilot
$100 (Max)
10,000 base plus 10,000 flex, 20,000 credits total, priority access to premium models. At the $0.01 rate that is $200 of metered value for $100.
Team
OpenAI Codex
$20 / user / mo
Business, 2 or more users, billed annually. $25 per user per month billed monthly. Adds SAML SSO, MFA, larger cloud VMs, and no training on your business data by default.
GitHub Copilot
$19 / seat / mo
Business, 1,900 credits per user per month, pooled across the enterprise. Self-serve sign-up has been paused for GitHub Free and GitHub Team orgs since April 22, 2026, so you go through sales.
Enterprise
OpenAI Codex
Contact sales
Enterprise and Edu add SCIM, EKM, RBAC, audit logs through a Compliance API, and data residency. With flexible pricing there are no fixed rate limits and usage scales with credits.
GitHub Copilot
$39 / seat / mo
Enterprise, 3,900 credits per user per month, priority model access. Published price, no sales call required to know what a seat costs.
No seat at all
OpenAI Codex
API pricing
Bring an API key and pay for the tokens Codex uses. Good for CI. No cloud features: no GitHub code review, no Slack.
GitHub Copilot
Not available
Copilot is sold per seat. There is no key-based mode, so an unattended pipeline needs a licensed seat behind it.
Two rungs deserve a second look. The bottom one, because Codex has an API-key mode and Copilot has nothing like it: an unattended pipeline on Copilot needs a licensed human seat behind it. And the team rung, because a $19 Copilot seat needs no annual commitment while the $20 Codex seat does. For the Copilot ladder on its own, in full, see GitHub Copilot pricing.
What a Codex plan actually gets you, in messages
This is the Codex equivalent of Copilot's credit allowance, and it is shaped nothing like it. OpenAI publishes local messages per rolling five-hour window, as a range, by model. The ranges are wide because a small script and a long refactor are both one message.
Swipe to see more
| Model | Plus $20 | Pro 5x $100 | Pro 20x $200 | Business |
|---|---|---|---|---|
| GPT-5.6 Sol | 10 to 100 | 50 to 500 | 200 to 2,000 | 10 to 100 |
| GPT-5.6 Terra | 25 to 200 | 125 to 1,000 | 500 to 4,000 | 25 to 200 |
| GPT-5.6 Luna | 250 to 2,000 | 1,250 to 10,000 | 5,000 to 40,000 | 250 to 2,000 |
| GPT-5.5 | 15 to 80 | 75 to 400 | 300 to 1,600 | 15 to 80 |
| GPT-5.4 | 20 to 100 | 100 to 500 | 400 to 2,000 | 20 to 100 |
| GPT-5.4 mini | 60 to 350 | 300 to 1,750 | 1,200 to 7,000 | 60 to 350 |
Three things about this table are easy to miss. The Business column is identical to Plus, so a team seat buys you admin controls and larger cloud machines rather than more throughput. Cloud chats draw from the same window as local messages, and OpenAI notes they "may use more of your allowance" because they run on Sol. And there are weekly limits on top that are not published here, so a good week can still end early.
Codex vs Copilot for a team of ten developers
The list prices land within $60 of each other, so the interesting column is the third one: whether you can tell your finance team what the month will cost before it happens.
Swipe to see more
| Plan, 10 people | Per month | Included | Forecastable? | Notes |
|---|---|---|---|---|
| Copilot Business | $190 | 19,000 pooled AI credits | Yes | Ten seats at $19. The credits pool across the enterprise, so the two people who live in the agent borrow headroom from the eight who mostly type. Past the pool it is $0.01 a credit and an admin can set a budget to stop it. |
| Copilot Enterprise | $390 | 39,000 pooled AI credits | Yes | Ten seats at $39, double the credits per head, priority model access, and the same $0.01 overage. Requires GitHub Enterprise Cloud. |
| Codex Business, annual | $200 | 25 to 200 Terra messages per user per 5 hours | No | Ten users at $20 on an annual commitment. The allowance is a rate, not a budget, so it cannot be exceeded and it cannot be converted into a dollar figure either. Buying extra workspace credits requires flexible pricing. |
| Codex Business, monthly | $250 | Same as above | No | The same product with no annual commitment costs 25 percent more. If you are comparing $19 Copilot seats against "$20 Codex", this is the number to compare against unless you are signing for a year. |
We have deliberately not put a dollar figure on Codex overage anywhere on this page. OpenAI publishes credits per million tokens and says GPT-5.6 "averages 5-30 credits per message", but no dollar price for a ChatGPT credit, so any monthly total for Codex would be a guess dressed as arithmetic. GitHub's $0.01 is a real number and that is why only one side of this table has a Yes in it. If you are sizing a Copilot rollout specifically, what GitHub Copilot costs for a team does the seat math, and GitHub Copilot Business covers the seat-buying rules.
You may not have to choose: Copilot already bills for Codex
"Can I use Codex with a Copilot subscription" is one of the most searched questions in this cluster and the comparisons in this SERP do not answer it. GitHub does, in its own billing table.
The billing row
GitHub lists the "OpenAI Codex Visual Studio Code integration" as a Copilot-billed feature: while in preview, each prompt to Codex uses one premium request multiplied by the model multiplier, charged to your Copilot allowance.
The plan gate
Footnote three on GitHub's plans page: GPT-5.4 nano "is currently only available in the Codex Visual Studio Code extension (Copilot Pro+ only)". So Pro+ at $39 is the tier that unlocks that model in that extension.
The shared models
Copilot's model list for Pro and above includes GPT-5.3-Codex and the GPT-5.6 Sol, Terra and Luna family. The same OpenAI models power both, so this was never a model comparison.
Be precise about the scope, because it is narrow. What GitHub bills is the Codex extension inside Visual Studio Code, in preview. It is not Codex cloud, not the Codex CLI, and not the @Codex pull request review, all of which still want a ChatGPT plan or an API key. The practical upshot for a Copilot Pro+ holder is that you can evaluate the Codex editor experience this week without buying a second subscription. Something similar is true in the other direction inside Cursor, which we cover in whether you can use Codex in Cursor.
Where each one runs, feature by feature
The rows that decide most evaluations are the first one and the GitLab one. Everything else, both products do.
Swipe to see more
| Surface | OpenAI Codex | GitHub Copilot |
|---|---|---|
| Inline code completion | Not offered | Yes, unlimited and unmetered on every paid plan |
| Chat in the IDE | Yes, through the IDE extension | Yes, plus inline chat, slash commands and chat skills |
| Terminal | Codex CLI, on any plan or an API key | Copilot CLI, included in every plan |
| Cloud agent on your repo | Codex cloud, on every ChatGPT plan, not on an API key | Copilot cloud agent, on every plan including Free |
| Pull request review on GitHub | Yes, tag @Codex or turn on automatic reviews | Yes, Copilot code review, at a published 13-credit multiplier |
| GitLab merge requests | Yes, in Beta, including merge request reviews | Not supported |
| Start work from Linear or Slack | Yes, both | Not part of the documented plan features |
| Desktop app | ChatGPT desktop app for local chats, API key included | Copilot app, in every plan |
| Mobile | Codex on iOS | Copilot Chat in GitHub Mobile |
| MCP servers | Yes, and each one you enable burns more of your allowance | Yes, on every plan including Free |
| Run it with no subscription | Yes, on an API key, at API token rates | No |
If your pull requests are on GitLab, this table has already decided it for you: Copilot code review is bound to GitHub pull requests and Codex reads GitLab merge requests in Beta. We keep a wider survey of that constraint in which AI coding agents work with GitLab. On the Copilot review side specifically, GitHub Copilot code review works through the 13-credit multiplier and the one limit that stops it being a merge gate.
The fine print on both sides that changes the decision
Each of these is quoted or paraphrased from the vendor's own documentation, read in August 2026. Four cut against Codex, three against Copilot, and the last one against both.
Codex
Your cloud tasks eat your local coding budget
OpenAI is explicit: "On ChatGPT plans, local messages and cloud chats share a five-hour window. Additional weekly limits may apply." So delegating three long jobs to Codex cloud in the morning can leave you rate-limited in your editor in the afternoon. There is one pot, not two, and a weekly ceiling you cannot see in the per-window table.
Codex
An Enterprise seat can be limited like a $20 seat
Straight from the pricing doc: "Enterprise and Edu plans without flexible pricing have the same per-seat usage limits as Plus for most features." If your procurement lands an Enterprise agreement without flexible pricing, your developers get Plus-level throughput. Ask which billing model the contract puts you on, in writing, before signing.
Codex
The best model is on promotional pricing with a date on it
The rate card carries the line "GPT-5.6 Sol's promotional pricing is available at least through November 21, 2026." Sol is the top reasoning model and the one cloud chats use. Its credit rate is 100 per million input tokens and 500 per million output today. Budget on the assumption that number is provisional.
Codex
Monthly billing costs 25 percent more on Business
Business is "$20 per user per month, 2+ users, billed annually. $25 per user per month when billed monthly." Almost every comparison quotes the $20 and skips the condition. Against a Copilot Business seat at $19 with no annual commitment required, the real month-to-month gap is $6 a head, not $1.
Copilot
You may not be able to buy Business without talking to sales
GitHub's own plans page carries this notice: "Starting April 22, 2026, new self-serve sign-ups for Copilot Business for organizations on GitHub Free and GitHub Team plans are temporarily paused." If you are a small team on GitHub Team expecting to add Copilot Business with a credit card this afternoon, budget for a sales cycle instead. Codex Business you can sign up for on the web.
Copilot
Not every action costs one credit
The consumption table is per feature, not flat. Chat is one request times the model multiplier. The cloud agent is one per session, plus one for every real-time steering comment you leave while it runs. Code review is a fixed 13. Spark is a fixed 4. A month of "a few prompts a day" is not the same bill as a month of agent sessions you keep nudging.
Copilot
Base credits are the only half you can budget on
The individual plans publish a base allowance and a flex allotment: Pro is 1,000 plus 500, Pro+ is 3,900 plus 3,100, Max is 10,000 plus 10,000. Base is fixed to the price of the plan. If you need a floor you can put in a spreadsheet without hoping, use the base figure and treat flex as upside.
Both
Neither one gives you a flat, predictable monthly bill
This is worth saying plainly because we sell against it. Copilot can bill you more than you planned. Codex can make you stop working. Both are usage-shaped products dressed as subscriptions, and if what you want is one number a month that never moves, neither of these two is that, whatever the sticker price says.
Codex or Copilot: which should you buy?
By situation, with the cases where the answer is neither of them left in.
Swipe to see more
| If this is you | Buy | Why |
|---|---|---|
| Most of your day is typing code and you want completions | Copilot | Codex has no autocomplete product. Copilot's completions are unlimited and never touch the credit meter on any paid plan, so the thing you use most is the thing you never pay per use for. This is not close. |
| Most of your day is handing off whole tasks and reading diffs | Codex | Codex cloud is built around parallel task environments you configure per repository, started from the web, GitHub, GitLab, Linear or Slack. Check the five-hour window for the model you plan to use first: on Plus that is 10 to 100 Sol messages, shared with your local editing. |
| Your repositories are on GitLab | Codex | Of these two, only Codex reads GitLab. It is in Beta and it does merge request reviews. Copilot code review is bound to GitHub pull requests and no amount of configuration points it elsewhere. |
| You need it running in CI with no human and no seat | Codex on an API key | OpenAI names this case: an API key is "great for automation in shared environments like CI", billed per token. You give up the cloud features. Copilot has no seatless mode, so an unattended pipeline needs a licensed seat behind it. |
| You already pay for ChatGPT Business or Plus | Try Codex first, it costs nothing new | Codex is included in the plan you already have, drawing on the allowance you already bought. Evaluate it for a fortnight before you add a second vendor. The same logic runs the other way for anyone already holding Copilot seats. |
| You are on Copilot Pro+ and want to try Codex in VS Code | You can, on your Copilot allowance | GitHub bills the OpenAI Codex VS Code integration as a Copilot feature, one premium request per prompt times the model multiplier, and GPT-5.4 nano in that extension is Copilot Pro+ only. This covers the extension, not Codex cloud or the CLI on a ChatGPT plan. |
| You are a small team on GitHub Free or GitHub Team | Check you can even buy Copilot Business | Self-serve sign-ups for Copilot Business have been paused for those org types since April 22, 2026. Find out whether sales will take a ten-seat order before you build a rollout plan around it. |
| Finance needs one predictable number every month | Neither, honestly | Both meter. One overspends, the other throttles. A flat per-seat product with no usage meter is a different shape of purchase, and that is the case we make for ourselves further down rather than pretending either of these does it. |
Weighing more than these two? GitHub Copilot against Claude Code and Codex against Claude Code add the third obvious contender, Cursor against GitHub Copilot covers the IDE-replacement route, and AI coding tool pricing puts every meter on one page.
Where we fit, and where both of them beat us
We sell a competing product, so take the losses first. Copilot has completions, and completions are the feature developers touch a hundred times a day; we do not sell them and neither does Codex. Copilot is already inside the platform your code lives on, one seat away, with org policy controls and audit logs we cannot match. Codex has the deeper bench of surfaces: parallel cloud environments per repository, Linear and Slack entry points, and an API-key mode that runs unattended in CI. If either of those descriptions is the job you are hiring for, buy theirs.
What we do differently is the shape of the bill and the shape of the output. Agentcode takes a task against a connected GitHub or GitLab repository, plans it, edits the code, runs your test suite and opens a pull request for a human to read. It never merges on its own, so your branch protection and your reviewers stay exactly as they are. It costs $29 a month, flat. There is no credit balance draining in the background, no five-hour window to wait out, and no month where the invoice surprises you because someone ran a big refactor on the 28th.
That is a narrower product than either of these. It is the right one if the thing you keep wanting is finished, reviewable work rather than faster typing, and if a predictable line item matters more to you than a bigger menu. Plenty of teams run one of these two alongside us and that is a sensible setup, not a compromise: see whether you should use more than one AI coding assistant.
Codex vs Copilot: the questions buyers actually ask
Can I use Codex with a Copilot subscription?
For the VS Code extension, yes. GitHub's billing table lists the OpenAI Codex Visual Studio Code integration as a Copilot-billed feature: while in preview, each Codex prompt uses one premium request times the model multiplier. GPT-5.4 nano in that extension is Copilot Pro+ only. Codex cloud and the Codex CLI still need a ChatGPT plan or an API key.
Can I use Codex CLI with a GitHub Copilot subscription?
No. What GitHub bills against your Copilot allowance is the Codex extension inside Visual Studio Code, not the standalone CLI. Codex CLI authenticates with a ChatGPT plan or an OpenAI API key. On a key you pay per token and lose the cloud features, which OpenAI recommends for CI.
Is Codex better than Copilot?
They answer different questions, so the honest answer is neither. Copilot is the only one of the two that does inline completion, unmetered. Codex is the stronger delegation tool, with parallel cloud environments and GitLab, Linear and Slack entry points. Pick on which half of your day you want help with.
Is Codex included in ChatGPT Plus?
Yes. A $20 Plus subscription includes Codex on the web, in the CLI, in the IDE extension and on iOS, plus cloud integrations like automatic code review and Slack. Usage runs against your plan's rolling five-hour window, published as 25 to 200 GPT-5.6 Terra messages or 10 to 100 on Sol.
Is Codex included in ChatGPT Business?
Yes, at $20 per user per month billed annually or $25 monthly, with a 2-user minimum. Business adds larger cloud VMs, SAML SSO, MFA and no training on your business data by default. The per-seat usage window is the same as Plus: 25 to 200 Terra messages per five hours.
Codex vs Copilot: which is cheaper for a team of ten?
On list price Copilot Business wins narrowly, at $190 a month against $200 for Codex Business annually or $250 monthly. But only the Copilot side can be forecast: 19,000 pooled credits and a published $0.01 overage. OpenAI publishes no dollar value for a ChatGPT credit, so a Codex overage cannot be priced in advance.
Does Copilot use the same models as Codex?
Partly, yes. GitHub's model list for Copilot Pro and above includes GPT-5.3-Codex and the GPT-5.6 Sol, Terra and Luna family, which are the same OpenAI models Codex runs, alongside Claude and Gemini options. So this comparison is about metering, surfaces and workflow, not about which model you get.
Does Codex have a free tier?
Yes. OpenAI lists a Free plan at $0 for exploring Codex on quick coding tasks, with Go at $8 above it. Copilot Free also exists, with an allowance of AI credits, auto model selection only and completions capped at 2,000 a month, and only for developers without an org licence.
Does Codex have limits?
Yes, and they are rate limits rather than budgets. OpenAI publishes local messages per rolling five-hour window by model and plan, notes that "additional weekly limits may apply", and that local messages and cloud chats share the same window. Plus and Business get the same per-seat figures.
Codex vs Copilot in VS Code: which should I install?
Both, if you have the licences, because they do different jobs in the editor. Copilot handles completions and inline chat. The Codex extension handles delegated work on a task. If you hold Copilot Pro+ you can run the Codex extension on your Copilot allowance rather than adding a ChatGPT plan.
What happens when you run out of Copilot AI credits?
Nothing stops. Usage past your allowance bills at $0.01 per AI credit, pooled across the enterprise on the org plans, and an admin can set a budget to cap it. Code completions and next edit suggestions are never billed in credits and stay unlimited on every paid plan.
Codex or Copilot for an enterprise rollout?
Copilot if you want a published per-seat price and GitHub-native controls: $39 an Enterprise seat, 3,900 credits, pooled. Codex if you need GitLab, Linear or Slack entry points, SCIM, EKM or data residency. Check whether your Codex contract includes flexible pricing, because without it seats are limited like Plus.
Last updated: August 2026
Keep researching
More on Copilot: GitHub Copilot pricing, GitHub Copilot Business, Business against Enterprise, the Copilot Max plan and is GitHub Copilot worth it.
More on Codex: Codex against Cursor, Codex and Claude Code usage limits and how to budget for AI coding tools.
One meters your credits. One meters your patience.
Connect a GitHub or GitLab repository, describe a task, and read the pull request it opens. Flat $29 a month, no credit balance, no five-hour window, and nothing merges without you.