GitHub Enterprise Adds Copilot, Bills From First Use

If you buy GitHub Enterprise Cloud through an account team, you can now give every licensed developer Copilot at no seat charge, and Chat, code review and cloud agent work bill from the first request. GitHub announced the change on October 28, in a Universe release that also made Copilot Fusion generally available.

At list prices, the bundle cannot cost a licensed user more than a Copilot Business or Enterprise seat. It does remove the prepaid floor, so your renewal question changes from how many seats to buy to how much AI usage to budget. Fusion's 30% discount runs through November 30, which means a budget built on November's Fusion usage will run low in December if the discount lapses.

---

What Does GitHub Enterprise Now Include?

GitHub Enterprise now includes Copilot access with the same feature and model eligibility as Copilot Enterprise, and no AI credits. GitHub's plans page sets out the options, which are separate from whatever you pay for your GitHub licences. GitHub's pricing page listed Enterprise at $21 per user per month when checked on November 2, 2026.

Option	Seat charge per month	Included AI credits per user
Copilot Business	$19	1,900 (pooled)
Copilot Enterprise	$39	3,900 (pooled)
Copilot access, GitHub Enterprise licence	$0	None
Copilot Standalone, no GitHub Enterprise licence	$10	None

One AI credit is $0.01, so the Business and Enterprise pools are worth exactly what those seats cost. GitHub's billing docs say the new access has "no included allowance" and that metered usage is billed from first use.

Availability is narrower than the headline. The changelog limits it today to GitHub Enterprise Cloud customers working with a GitHub or Microsoft account team. Self-serve opens in stages, with dates promised "in the coming weeks", and current self-serve subscriptions are unaffected. Business and Enterprise remain on sale. The plans page also says Copilot is not currently available for GitHub Enterprise Server, so a self-hosted estate gets nothing from this until that changes.

Does the Bundle Save Money Compared With a Copilot Seat?

For an enterprise whose users all hold GitHub Enterprise licences, the bundle costs the same as seats or less at list price, because a seat's only metered advantage is credits worth what you paid for it. You save only when total usage falls short of the pool, since heavy users already draw on whatever light users leave unspent. The choice also covers the whole enterprise: GitHub's transition doc says you cannot combine the bundle with Business or Enterprise seats, including in individual organizations, so you cannot move only your light users. Standalone is different: its $10 buys no usage, so it beats Business only below about $9 of average monthly usage and Enterprise below about $29.

Here is that arithmetic for 100 developers, using the list prices above and assuming the pool is shared evenly. It covers the Copilot line only. The GitHub licence is separate, and a negotiated discount changes every cell.

Average usage per developer	Business	Enterprise	Bundle	Standalone
$12 a month	$1,900	$3,900	$1,200	$2,200
$30 a month	$3,000	$3,900	$3,000	$4,000
$60 a month	$6,000	$6,000	$6,000	$7,000

The strongest case for keeping seats is forecasting. A seat gives finance a known minimum, and your account team may have priced it below list, which this table cannot see. If your Copilot discount is deep, a $0 seat charge may be worth less to you than the list comparison suggests, so get both quotes in writing.

There is also a real gain at the bottom of the range. Inline completions and next edit suggestions are not billed in credits and stay unlimited, even at a $0 budget. A developer who only wants completions now costs you nothing beyond the licence you already pay for.

---

Budgets Stop Spending Only If You Configure Them To

With no pool to absorb early usage, the budget becomes your only cost control, and an enterprise spending limit does not stop anything until you enable "Stop usage when budget limit is reached". GitHub's docs offer budgets at the user, cost center, organization and enterprise levels, set in US dollars. A user-level budget is a hard stop. The changelog adds that Chat, code review and Copilot in the cloud need available budget, while running out does not remove a user's access.

GitHub gives no budget figure to start from, and the migration guidance states no default amount. It does warn that agent mode and the cloud agent make several model calls per task. Your own usage dashboard is the best source you have.

Code review is the line most likely to surprise you, because admins can set an effort level as a default for a whole enterprise. GitHub's pricing page for models estimates an average review at about $0.25 on Lite, $1.50 on Balanced and $10 on Max, with Max typically ranging from $1.50 to $25 and GitHub Actions minutes excluded. The model is chosen automatically and not disclosed.

Take 100 developers merging 1,000 pull requests a month, an illustrative volume you should replace with your own. Lite averages about $250, Balanced about $1,500 and Max about $10,000, or $100 per developer. That one default can outrun the rest of the Copilot bill. Max is in public preview, and admins can cap the highest effort level available.

Copilot Fusion Is GA, and Its Discount Ends November 30

Copilot Fusion is a model picker option that chooses the models and orchestration for each task, and you pay for every model it calls at that model's standard rate. The GA announcement describes three workflows. Single uses one model. Cascade has an efficient model draft and escalates if a quality check fails. Critique has one model draft, another review and the first revise. We worked through where a cascade stops saving money when this shipped as the HydraFusion preview, and that break-even logic still applies.

The docs say there is no separate Fusion charge, and that paid plans get 30% off model costs when using it through November 30, 2026. The page does not say what follows. If the discount simply lapses and nothing else changes, the same tokens cost about 43% more on December 1 (1 divided by 0.7). Two models on GitHub's rate card carry their own promotional clocks as well: Gemini 4 Argon until December 1 and Gemini 3.8 Flash through December 31. GitHub's Fusion docs do not list the model pool or name Gemini, so you cannot tell from them whether either clock touches your Fusion bill.

GitHub's headline claim is that Fusion Max scored 60 on the Artificial Analysis Coding Agent Index, within two points of Claude Code with Opus 5.5 Max, at about half the average cost per task, about half the tokens and about 40% less time. Artificial Analysis's own write-up backs the cost and token comparison for Fusion's max setting in Copilot CLI: $5.59 per task against $11.42, and 7.2M tokens per task. It gives no time figure, so the 40% rests on GitHub's word. The same page lists Devin Fusion at 62 for $8.00 per task, and has Fusion leading Terminal-Bench 4.0 at 71 while trailing Claude Code on FrontierCode, 51 to 56. The index measures pay-per-token API cost across three benchmarks, and this result covers only the max setting, so it will not predict your task mix or what Fusion Standard costs you. The HydraFusion preview's savings moved a long way between benchmarks too, which is why the test that counts is your own repositories.

Rollout may also be uneven. In the launch discussion, a Copilot Business administrator reported on October 30 finding neither Fusion nor a setting for it, and a Pro user said on November 2 that it was still missing from VS Code. Those are two reports with no staff reply visible yet, but confirm your own access before you plan a November test around the discount.

---

Copilot Approvals Can Now Satisfy Required Reviews

Copilot's approval can now count toward a repository's required approvals, a generally available setting that is off by default. Per the code review changelog, enterprise, organization or repository admins can turn it on, repository admins can choose which file paths Copilot may approve, and new commits dismiss the approval.

This is the release's real governance decision, and your auditors may have a view on it. GitHub's coding agent review doc says your own approval of a Copilot pull request won't count and "another reviewer must approve". It does not say whether Copilot itself can be that other reviewer on a pull request Copilot wrote. Test that in a sandbox repository before enabling anything. We have covered the independence problem when one vendor writes and reviews the code, and what Zalando saw after auto-approving pull requests.

The Rest of the Universe Release Runs on the Same Meter

Most of what else GitHub shipped can consume AI credits, and it now reaches more of your repositories. Each of these needs an owner:

• GitHub Agentic Workflows is generally available. Agents run read-only by default in a sandboxed container behind a firewall, and gh aw forecast estimates monthly AI credit spend. The default engine requires a Copilot licence. GitHub's post also carries a customer quote from Accenture claiming AI writes 91% of one team's commits, which is the customer's own number.
• Copilot Automations gained REST API triggers, 24 new GitHub events and chaining, plus AI credit limits per run, day, week or month. Automations without limits run as before, so existing ones stay uncapped until someone sets a limit.
• Agent Merge lets Copilot fix failing workflows and merge conflicts in the cloud, and can combine with auto-merge. The post carries no GA or preview label. We have written before about what retry loops did during a GitHub outage, and an agent that responds to failing checks deserves the same cap.

Separately, the GitHub Copilot SDK is becoming the Microsoft Copilot SDK, with a planned move to the Microsoft organization and no date given. GitHub says existing users need to change nothing now. If you have built on it, note the move and check that your support assumptions still hold.

---

What to Do Before November 30

This Week:

1. Pull 90 days of GitHub Copilot usage per developer from your dashboard and total it against your seat pool (seats times $19 or $39). The bundle saves money only when total usage falls short of that pool.
2. Ask your account team for a written quote on the bundle against your current seat price, including what happens to any existing Copilot commitment.
3. Check every organization for code review defaults and cap the maximum effort at Balanced until you have measured Max.

This Month:

1. Run Fusion Standard against your current default on real tasks in two repositories, and record credits per merged pull request. Calculate the result at both the discounted rate and the rate 43% higher.
2. Decide who owns the enterprise budget, then enable "Stop usage when budget limit is reached" and set user-level budgets for the heaviest agent users.
3. Set limits on every existing Copilot Automation, starting with anything triggered by an external system.

Before Renewal:

1. Model three budgets from your usage data (current, double and five times) and take the highest one you could defend to finance into the negotiation.
2. Get your security or audit lead to write down a decision on Copilot approvals, by file path, before any admin enables them.

The Bottom Line

GitHub has moved the same way Microsoft did when it billed Copilot Autopilot outside the $30 seat: access gets cheaper or free, and agent work goes on a meter. We argued in our true cost of a Copilot seat analysis that the licence was only the floor. GitHub Enterprise customers can now have that floor set to $0, with the variable part left for them to forecast.

For a licensed enterprise whose total usage sits below its seat pool, take the bundle, because the list math favours it. For a heavy agent shop with a deep seat discount, the case is much weaker, and you should compare quotes first. Either way, write the budget number down before November 30, using the undiscounted rate.

Continue Reading

• GitHub's Router Bills Every Leg. Find Your Break-Even.
• True Cost of a Copilot Seat: The Licence Is the Floor
• Best AI Coding Assistant at 500 Seats: Buy Copilot Business
• Same Vendor Wrote and Reviewed 208,145 PRs. Split Them.
• Agentic AI Pricing: Don't Buy Consumption Without a Cap
• Cursor Alternatives That Survive Security Review Start With Copilot