GitHub Copilot Now Bundled in Enterprise With Usage-Only Billing

What GitHub changed

GitHub is dropping the separate Copilot subscription for some GitHub Enterprise customers. Customers who work with a GitHub or Microsoft account team can now get Copilot included for every user who holds a GitHub Enterprise license. They pay for AI usage instead of buying a Copilot seat for each person.11 The company announced the change at its Universe conference in San Francisco on October 28, along with a new Copilot app, a model router called Copilot Fusion, and a deeper code review tier.15

The billing details matter more than the bundle itself. Usage is metered from the very first request and charged against the enterprise's AI budget. There is no included allowance to track or top up.11 People inside an enterprise account who don't hold a GitHub Enterprise license can get the same Copilot access through a new Copilot Standalone option for $10 per user per month.11 Both routes come with the same features and models as Copilot Enterprise, but neither includes monthly AI credits per seat.11

GitHub also limited what happens when the money runs out. Code completion and next edit suggestions keep working with an empty budget. Copilot Chat, Copilot code review and Copilot in the cloud stop until more budget is available.11 The company says an exhausted budget never takes away the seat itself.15

This isn't available to everyone yet. Account-team customers on GitHub Enterprise Cloud can switch now, self-serve customers will get it in stages, and existing self-serve subscriptions are unaffected.11 Copilot Business and Copilot Enterprise are still on sale, and current subscriptions don't change.15

The bundle finishes a change that started in April

The move makes more sense against GitHub's pricing history this year. On April 27, GitHub said every Copilot plan would switch to usage-based billing on June 1. The old "premium request" was replaced by the GitHub AI credit. Each credit is worth one cent and is used up by input, output and cached tokens at each model's published rate.2 Seat prices stayed at $19 for Business and $39 for Enterprise. Each seat adds 1,900 or 3,900 credits to a shared pool, and anything beyond the pool costs $0.01 per credit.7

The last step before this week came on September 1, when the transition promotion ended. Business seats went from 3,000 to 1,900 credits a month, and Enterprise seats from 7,000 to 3,900. That's a cut of about 37% and 44%.18 One cost-analysis firm noted that because the summer pool was subsidized, many enterprise admins wouldn't see what their real usage costs until the October invoices.12

The new option goes further: it removes the seat fee altogether and keeps only the usage meter. One analyst summed it up this way: what drives the developer-tools bill is moving from how many people you employ to how much work your agents do.13

Who comes out ahead

The clearest winners are large organizations where adoption is uneven. A $39 seat is hard to justify for a developer who barely uses Copilot. Under usage-only billing, that person costs only what they actually consume.13 There is a hidden cost in the old setup too. Copilot Enterprise requires GitHub Enterprise Cloud at roughly $21 per user, so the real per-developer price was closer to $60 before any extra usage.3 For companies already paying for GitHub Enterprise licenses, the bundle makes Copilot access a feature of the platform rather than a second product to buy.

The trade-off is predictability. Even under the seat model, extra usage is billed by default unless an admin turns it off.18 Agentic sessions can also use up credits at a rate no seat count would predict. One analysis described a single heavy session burning nearly half of an Enterprise seat's monthly allotment.12 With no included pool, every token in the new model is billable from the start. That makes the budget controls, alerts and cost attribution GitHub points to essential rather than optional.11 Heavy agent users won't get cheaper bills from this change. It mainly stops companies from paying for light users who rarely touch Copilot.

The coverage mostly agrees on the mechanics, but the framing differs. GitHub's Universe post presents the change broadly: Enterprise customers can use all Copilot features without a separate subscription.31 The changelog is narrower. It limits availability to account-team customers on Enterprise Cloud and spells out which features need budget.11 Some third-party pricing guides are also out of date. One still lists Copilot Enterprise as including 1,000 premium requests per seat, a unit GitHub retired on June 1.5 Buyers should rely on GitHub's own documentation.

Agents are why the pricing changed

GitHub's Octoverse data, released the same day, helps explain why per-seat pricing stopped making sense. More than 40% of pull requests in September started as a draft written by an agent, up from 17% in March.32 Among Copilot users who chose their own model, 61% used more than one.32 When the work is done by agents running across days and many models, the number of human seats is a weak measure of cost.

That load has hurt GitHub's reliability. Microsoft CoreAI chief Jay Parikh admitted the platform struggled at times this year. One analysis put GitHub's 90-day availability at about 84% in early May, based on reconstructed status-page data.34 GitHub says it has added 4 million CPU cores, 160 PB of storage and 50 Tb/s of network capacity this year. It expects to finish moving to a multi-region Azure deployment early next year.32 Usage-based billing lets GitHub charge for that growing, round-the-clock load as it happens.13

Copilot Fusion is the cost-control piece. It decides which model, or mix of models, should handle each step of a task.32 GitHub says Fusion Max scored within two points of Anthropic's Claude Opus 5.5 Max on the Artificial Analysis Coding Agent Index, at half the cost and 40% less time per task.33 Satya Nadella went further, claiming cost savings of up to 77%.32 These are GitHub's own benchmark claims, not independent results.34 GitHub is also offering a 30% discount on Fusion usage through the end of November. For customers paying per token, a router that picks cheaper models is effectively a price cut.

Code review is now the main battleground

AI code review gets particular weight in this launch. GitHub introduced Copilot code review Max, a deeper tier that it says posted the highest F1 score on ReviewBench for high-severity issues. The existing Lite and Balanced tiers remain for routine reviews.31 GitHub says its code review feature has produced 200 million findings since it launched last year.

This is where the billing model has teeth. Copilot code review is one of the features that stops working when the budget runs out.11 So a heavier review tier means more spending, not just better coverage. Teams should test Max on their highest-risk repositories first. The question is whether it catches bugs their own reviewers would miss, or just adds noise for humans to sort through.34

Anthropic got there first on several fronts. Claude Code's Code Review feature, which reviews an organization's pull requests, became generally available on October 15.27 Version 2.1.304 added review levels that can be set per repository.30 An earlier release, 2.1.288, added a --max-findings option to /code-review.30 Both companies now offer tiered review, so engineering teams choose how deep, and how expensive, each review should be.

How this fits with Claude Code and other tools

The wider AI coding market is moving quickly. In mid-October, Claude Code, Codex CLI, OpenCode and VS Code all shipped updates within a few days of each other, each aimed at running more agents on their own.22 Claude Code now turns on forked subagents by default.22 It also uses Opus 5.5 as its default model, priced at $4 per million input tokens and $20 per million output tokens.26

Pricing models still differ. Claude Code comes with Claude Pro and Max subscriptions and limits usage over five-hour windows plus weekly caps. Cursor bills a credit pool at the list price of whichever model handles the work.3 Many enterprises run Copilot alongside Claude Code or Cursor, each with its own billing that the others' dashboards can't see.12 Copilot Business and Enterprise also let organizations allow third-party agents such as Claude and Codex through org policy.4 Those agents draw from the same pool of GitHub credits.18

That is GitHub's strategic bet. If Copilot comes with the platform at no seat cost, the only thing a competitor can win is usage, and GitHub can bill third-party agents through its own credits too. Bundling Copilot into Enterprise makes GitHub the place where AI coding gets billed, whichever model or vendor does the work. For buyers, the September invoice is the best starting point for judging the new model. The more important step is to set firm budgets before agent activity grows further.