GitHub Copilot Pricing: AI Credits, the Shared Pool, and What Actually Bills

The seven plans and what each one includes

GitHub Copilot sells in seven tiers. Five are individual, two are for organisations. Every paid plan carries a monthly allowance of GitHub AI Credits, which is the unit GitHub meters Copilot usage in.

Two details in that table are easy to skim past. The individual plans split their allowance into a base figure and a flex allotment, and GitHub describes flex allotments as variable, so the headline total is not a fixed entitlement. And Copilot Free and Copilot Student carry an allowance GitHub does not put a number on, which means neither can be modelled.

What a credit bills, and what it does not

This is the correction most teams need first. Code completions, the thing most people picture when they hear "Copilot", do not consume credits at all. GitHub's documentation states that code completions and next edit suggestions are not billed in AI credits and remain unlimited on all paid plans.

What does draw on the allowance, per GitHub, is chat, agent mode, code review, the Copilot cloud agent and Copilot CLI. So the feature most people picture when they hear Copilot is unmetered, and the features teams have been moving toward are the metered ones. A team whose usage looks flat on a completions dashboard can be running a rising bill underneath it.

The pooling rule, which is the one that surprises people

For organisations, credits are not a per-user cap. GitHub's own wording is that each licence contributes AI credits to a shared enterprise pool, and usage beyond the pool is charged at $0.01 per AI credit.

Read that as a budgeting fact rather than a billing detail. A hundred Copilot Business seats do not buy a hundred separate 1,900-credit buckets that each stop when they are empty. They buy one pool of 190,000 credits that anybody can draw from. Three developers running agents across a large monorepo all month are drawing down the same pool that the other ninety-seven seats contributed to. There is no per-user stop in that design, so the control is GitHub's usage report rather than a limit that fires on its own.

That is not a criticism of pooling, which is usually the right design and is kinder than hard per-user caps. It is a statement about where your controls have to live. Pooled consumption needs a budget and a monitoring cadence, because the mechanism that would otherwise stop it does not exist.

The overage is on by default

GitHub states plainly that by default, users can continue past their plan's monthly allowance, and that each extra request is charged to the enterprise. There is no stop at the line. The allowance is a billing boundary, not a usage one.

GitHub itself recommends setting up budgets on its metered products, and singles out customers who had not previously enabled paid usage for premium requests. That is a fair warning and it is worth acting on. The documented behaviour is simply that usage continues past the allowance until a budget is configured.

The break-even GitHub publishes, and the unit problem underneath it

GitHub gives enterprises a genuinely useful number. On its page for choosing an enterprise plan, it says Copilot Business users who use more than around 800 premium requests a month would save money on a Copilot Enterprise plan. It also says to treat the Business allowance as a baseline rather than a limit, and notes that teams using agent mode, the coding agent and code review are likely to pass it.

Here is the problem with acting on that number today. GitHub's documentation is currently split across two billing vocabularies. The plans page and the organisation billing pages measure in AI Credits. The page an enterprise buyer is sent to in order to choose a plan still describes allowances, defaults and that break-even in premium requests. Both were live on 18 October 2026.

So the one actionable threshold GitHub publishes for the Business-versus-Enterprise decision is denominated in a unit its credit-based billing pages do not report. If your usage report comes back in credits, 800 premium requests is not a line you can check yourself against, because GitHub publishes no conversion between the two on either page. Treat the figure as directional, and make the decision on your own measured consumption rather than on the threshold.

You may now be running two Copilot credit systems

If your organisation buys both Microsoft 365 Copilot and GitHub Copilot, you have two separately metered currencies with nearly the same name, and they are not interchangeable.

Microsoft 365 Copilot meters agent usage in Copilot Credits, at $0.01 per credit pay-as-you-go, with prepaid capacity packs of 25,000 credits for $200, which works out at $0.008 per credit. GitHub Copilot meters in GitHub AI Credits, at $0.01 per credit beyond the pool. The same pay-as-you-go rate, a cheaper prepaid rate on the Microsoft side, and otherwise a different product, pool, admin surface and report.

The practical consequence is a reporting one. A finance question like "what did Copilot cost us last month" has two answers that live in two places and cannot be added up from one dashboard. Decide now which function owns each, because the alternative is discovering the split during a budget review.

What to do this week

Four things, in order, and none of them takes long.

Set a budget before anything else. It is the control that caps pooled spending automatically, and it is off until somebody sets it. GitHub also exposes a separate paid-usage policy for requests beyond the allowance, so check both rather than assuming one covers the other.

Pull the usage report. GitHub provides one that shows consumption per developer. You cannot make the Business-versus-Enterprise decision without it, and the published break-even will not substitute for it given the unit split above.

Separate completions from metered usage in your own reporting. Completions are included and unlimited on paid plans; the metered features are where overage comes from. Any adoption metric that mixes them will tell you the wrong thing about both.

Name an owner for each credit system. If you run both Copilots, write down who watches which pool. Two metered systems with the same word in the name is exactly the shape of problem that ends up owned by nobody.

Sources

Every price, allowance and quoted rule on this page was read on 18 October 2026 from GitHub's own documentation: the plans page for the tiers, prices and credit allowances, and the enterprise plan guidance for the default overage behaviour and the 800-request break-even. The Microsoft 365 Copilot Credit figures come from our own Copilot Credits explainer. Prices change. Check both pages before you commit a budget to them.