Out of Copilot credits mid-month: what a developer can do

Somewhere in the middle of the month, with work open in your editor, GitHub Copilot either stopped answering or quietly started costing your company money. Which of the two happened was decided by settings on the billing side of your organization, and you probably never saw them.

This page is written for the developer on a Copilot Business or Enterprise seat who just hit that wall. It covers what actually happened, what you can see about your own position, what you can ask for before the pool resets, and what resets on the first of the month. The page for the person who owns the bill is the premium-requests explainer.

What actually happened

GitHub Copilot moved organizations and enterprises to usage-based billing in June 2026¹. Since then, premium work has drawn on a monthly allowance of AI credits that is pooled at the billing entity rather than held per seat, so 100 Business users share one pool of 190,000 credits². Copilot Chat, the Copilot CLI, the cloud agent, Copilot Spaces, Spark, and third-party coding agents all consume AI credits². Code completions and next edit suggestions are not billed in AI credits and remain unlimited on paid plans², so a wall that stops chat and agents leaves autocomplete running. Consumption is metered on tokens, priced by the model used, and converted at 1 AI credit = $0.01 USD².

On September 1, 2026 the pool shrank. The promotional monthly allotments of $30 per Business user and $70 per Enterprise user, worth 3,000 and 7,000 credits at 1 AI credit = $0.01², ended on that date¹. The standard allotments are 1,900 credits per Business user and 3,900 per Enterprise user each month². Consumption that fit comfortably inside the pool through August stopped fitting in September.

Whether you stopped or your work started billing was decided by two settings. The first is the AI credits paid usage policy, which is enabled by default². With it enabled, consumption past the pool continues and bills at each model’s published rate. With it disabled, usage is blocked until the next billing cycle, when the monthly amounts refresh². The second is a user-level budget. When you exhaust one, your Copilot access is halted even if the organization’s pool still has credits, and a budget of $0 blocks you immediately². There is no automatic fallback to a cheaper model at either limit². Work stops rather than degrading.

The pool also empties faster than most of the people sharing it expect. Roughly 20% of users draw 70–80% of the credits³, and in a shared pool they set the date everyone else runs out. One developer reported the included credits nearly gone on day 11 of the month, with lighter usage than the month before⁴. Hitting the wall in the second week says very little about your own consumption.

What you can see from your seat

Your own meter lives in Copilot settings. Open your profile picture on GitHub, choose Copilot settings, and read the Usage this cycle card. If a user-level budget applies to you, the card shows your consumption against it, in the form 450 / 1,000 AI credits used. Without one, it shows your total for the cycle⁵.

The same figures are reachable without leaving the editor. In VS Code, the Copilot icon in the status bar shows the features in your plan, your progress toward its limits, and when the allowance resets. In Visual Studio it is the Copilot icon and then Copilot Consumptions, and in JetBrains IDEs it is the Copilot icon and then View quota usage⁵.

What the meter will not give you is a line-item account. Consumed credits generally cannot be traced to specific prompts, sessions, or exact features at a per-request level⁵.

“The problem is that the cost is not inspectable.” GitHub Community, June 2026⁴

The pool itself is read on the billing side. The usage metrics API reports consumption at the enterprise, organization, repository, and user levels, and the usage dashboard visualizes the trends, with a given day’s data available within two full days⁶. Viewing organization Copilot metrics is a granted permission, including through an organization custom role⁶. Your card and your IDE views are the developer’s window. The question of who drained the pool, and how fast, is answered with tools your administrator holds.

What you can do before the reset

Most of the levers sit with the people who run the billing entity. Every one of those can be asked for, and the last one on this list is already yours.

1. Read your own meter first. Open the Usage this cycle card and note the figure⁵. Whether you are at 100% of a 1,000-credit budget or at 2% of a drained shared pool changes what you should ask for, and the card is the difference between a request and a complaint.
2. Ask the bill owner for a per-user budget change. Budgets are created and edited by enterprise and organization owners and billing managers⁷. A budget scoped to a specific user overrides the universal and cost-center budgets for that user⁷, so a raise for you does not require a raise for everyone. Members can ask for an increase when they exhaust a Copilot AI credits budget⁷, and when your credits run out, a banner on GitHub.com offers to request more budget from your administrator⁵. Bring the number from your usage card to that conversation.
3. Request the free usage report, so the pool is visible. The hardest part of the budget conversation is that nobody in it can see the pool’s shape without assembling it from exports. A free, read-only report of the organization’s Copilot pool is linked at the end of this page. An organization owner installs it as a GitHub App, and organization members can request the installation themselves. The request reaches the owner by email, and the owner decides⁸.
4. Send the administrator the controls page. The per-user caps, the paid usage policy, and the scoped budgets each do one thing and stop somewhere. We wrote up the budget-controls page for the person choosing where those caps sit this month.
5. Pick a less expensive model where the work allows it. Each model carries its own published per-token rate⁹, and your consumption converts to credits at the rate of the model you selected². The model picker is the one lever that is already on your side of the permission line.

What resets on the first

The pool returns to its full monthly amount at 12:00 AM UTC on the first day of each calendar month². That date is fixed. It does not move when licenses are added, removed, or billed². Included credits do not carry over between months², so whatever the pool did not use is gone at the same moment you get yours back. If the paid usage policy is disabled, the block on usage also ends at the reset, when the monthly amounts refresh².

The reset fixes nothing on its own. The same pool, the same distribution of heavy and light users, and the same settings produce the same wall at roughly the same date the following month.

The part that is out of your hands

Every setting in this story is a ceiling. The paid usage policy chooses between an open invoice and a hard stop for the whole billing entity. A scoped budget stops a group together. A per-user budget stops one person. Each of them can hold spending down, and none of them moves allowance from the developer who will not use it to the one who ran out. Both failures land in the same pool in the same month: the heavy cohort hits the wall³ while the light cohort’s remainder is forfeited at the reset². If the controls themselves are the conversation at your company, what each control does and where it stops is written for the administrator choosing a cap height.

We are building Spillway Ops for the route that is missing from that figure: allocation that would otherwise be forfeited flows to the developers who ran out, read from usage metadata and applied through the vendor’s own admin APIs, with no repository access, no prompt contents, and no code. If the wall found you this month, we would genuinely like to hear what it interrupted.

Before any of those conversations, it helps to know where the pool stands today. An organization owner installs the report below, and a member can start that installation request on GitHub themselves⁸.