GitHub Copilot Business and Enterprise AI credits cut on September 1: the pool math

Last verified: September 25, 2026, against GitHub's billing docs and plans page.

This is for org admins and developers on Copilot Business or Enterprise seats. On Pro, Pro+, or Max? The June-August promotion only covered Business and Enterprise, so this change is org-only; our AI credits math piece covers those.

Below is what changed, the pool math for a 50-seat org, and what to do if your org is now closer to the edge of its included usage.

What changed for GitHub Copilot Business and Enterprise AI credits on September 1?

On April 27, 2026, GitHub CPO Mario Rodriguez announced that existing Copilot Business and Enterprise customers would get promotional included usage for June, July, and August: $30 per user per month in AI credits for Business, and $70 for Enterprise. At $0.01 per credit, that is 3,000 and 7,000 credits. GitHub offered it for those three months only.

August was the last promo month. From September 1, the standard included usage applies: 1,900 credits per user per month for Business and 3,900 for Enterprise, per GitHub's docs. If your org budgeted or set expectations around the June-August numbers, this is the line where those numbers stopped applying.

Seat prices did not change: $19 per user per month for Business, $39 for Enterprise. The included credits now equal the seat price in dollars: $19 buys 1,900 credits, $39 buys 3,900. That is a clean mental model going forward: whatever you pay per seat in dollars, you get back as included credits (100 credits per dollar), per seat, per month. A public GitHub community discussion (#198015) exists on the change; we are not citing vote or comment counts because we have not verified them.

Before and after, per user

Business per-user credits fell by more than a third, Enterprise by over two-fifths, and both tiers now include credits worth exactly the seat price in dollars.

How does the Copilot credit pool work for an organization?

One AI credit equals $0.01. Tokens (input, output, and cached) are priced per model and converted to credits; for the per-model rate table, see our AI credits math piece, which we will not repeat here.

Credits pool at the billing entity, not per person. GitHub's own example: 100 Business users share 190,000 credits. Adding licenses mid-cycle grows the pool immediately. Removing licenses does not shrink it until the next billing cycle. In practice, this means a single heavy user can draw down the pool for the whole org, and there is no per-seat cap unless an admin sets one with budgets.

There is no rollover. The pool resets at 00:00:00 UTC on the 1st of each month. Credits are billed for Copilot Chat, Copilot CLI, the Copilot cloud agent, Copilot Spaces, Spark, and third-party coding agents.

What still does not use credits?

Code completions and next edit suggestions are not billed in credits. They stay unlimited on every paid plan, before and after September 1.

Worked example: a 50-seat Business org

This is an illustrative example, not a typical org. A 50-seat Business org pooled 50 x 3,000 = 150,000 credits per month during June through August. From September 1, that pool is 50 x 1,900 = 95,000 credits per month. The difference is 55,000 credits, or $550 per month of included usage that is no longer there.

Same arithmetic for Enterprise: per seat, credits went from 7,000 to 3,900, so multiply seats by 3,900 to get the new monthly pool. A 50-seat Enterprise org went from 350,000 credits per month to 195,000, a drop of 155,000 credits, or $1,550 per month of included usage.

How long 95,000 credits (or 195,000, for the Enterprise example) lasts depends on which models and features your team uses; the pool is the same size either way. We do not have usage-rate data to say how fast any given team burns through it, and we are not going to guess at a burn rate we cannot source.

What happens when the Copilot credit pool runs out?

If additional usage is allowed, usage continues at published per-credit rates and bills to the org, with the option to cap it with budgets. If additional usage is not allowed, Copilot access that needs credits is blocked until the next cycle.

GitHub's docs put it plainly:

“Additional usage is enabled by default for organizations and enterprises.”

In other words, overage is opt-out, not opt-in. Separately, a user-level budget that runs out halts that user's Copilot even if the org pool still has credits, with no automatic fallback to cheaper models.

How to turn off Copilot overage or cap it

1. Disable the AI credits paid usage policy in AI Controls to block usage past the included pool.
2. Or keep it on and set budgets in USD: user-level, cost-center, org-level, and enterprise spending limits.
3. A $0 user budget blocks that user immediately.

Did Enterprise lose more than Business?

Yes, in both absolute and percentage terms: Enterprise lost 3,100 credits per user (44.3%) against Business's 1,100 (36.7%).

During the promo, Enterprise got about 179 credits per seat dollar (7,000 / 39) versus about 158 for Business (3,000 / 19). From September 1, both get exactly 100 credits per seat dollar. The promo favored Enterprise more, so Enterprise loses more in the correction.

Which tier is the better deal is out of scope here: the choice between tiers turns on features and governance, not on credits-per-dollar, which is now identical.

What are your options if the pool runs out early?

A neutral list, roughly in order of how much control they hand you:

1. Cap it. Budgets, or disabling the paid usage policy. No extra cost; users are blocked when the pool or their budget is gone.
2. Pay the overage. Keep the default and budget for it at published per-credit rates.
3. Route heavy chat and agent work through a BYOK tool that calls a model provider directly. You pay the provider's per-token rates instead of drawing GitHub credits. Not free.
4. Run a local model for work it can handle. No per-token bill, but it needs capable hardware (VRAM) plus electricity, and local models are not a drop-in for every task. See /local-llms and how to run a local model with Ollama.

Bodega One Code is one example of options 3 and 4: a local-first desktop IDE with BYOLLM (10+ provider presets, or local models via Ollama or LM Studio). It is free for everyone during beta, commercial use included, and has no credit pool. Cloud models still bill at your provider's rates, and local models still need the hardware. For the full comparison, see our GitHub Copilot alternatives page.

Code completions stay unlimited on Copilot regardless of this change, so teams using Copilot mainly for completions are not affected by the September 1 cut.

Sources

• GitHub Copilot is moving to usage-based billing (April 27, 2026 announcement, Mario Rodriguez): github.blog/news-insights/company-news/github-copilot-is-moving-to-usage-based-billing
• Usage-based billing for organizations and enterprises (included credits, pooling, overage, budgets): docs.github.com/en/copilot/concepts/billing/usage-based-billing-for-organizations-and-enterprises
• GitHub Copilot plans and seat pricing: github.com/features/copilot/plans