Your Copilot seat price didn’t change. What’s inside the seat did.
GitHub Copilot Business still costs $19 per user per month in September. The AI credits included in that seat drop from $30 back to $19 — and from 1 October every assigned seat bills upfront, opened or not. A budget built on the seat price is now measuring the wrong number. What changed, cited to GitHub’s own posts.
Your GitHub Copilot seat costs the same this month as it did last month. What the seat contains does not. Copilot Business stays at $19 per user per month and Enterprise at $39, while the promotional AI-credit allowance behind those seats — $30 and $70 for June, July and August 2026 — reverts to the standard $19 and $39 from September. Nothing on the invoice line moves, so nothing trips a renewal review, and the usage you get for the money falls by more than a third.
This is what "moving beyond per-seat pricing" costs in practice. It is worth understanding as a mechanism rather than as one vendor's changelog, because the mechanism is spreading and it defeats the way most companies budget for software.
What changed in GitHub Copilot billing in September and October 2026?
Three dated changes, each from GitHub's own posts:
- From 1 September 2026, the included AI credits revert. GitHub's April 2026 pricing announcement set the standard rates — Copilot Business at "$19/user/month, including $19 in monthly AI Credits" and Enterprise at "$39/user/month, including $39 in monthly AI Credits" — and granted existing customers promotional credits of $30 and $70 "for June, July, and August." That window has closed. The seat price is unchanged.
- From 1 October 2026, assigned seats bill upfront. The 28 August 2026 changelog states that "all Copilot Business and Copilot Enterprise seats assigned will incur an upfront charge." Payment comes before access.
- From 28 September 2026, chat data is retained for the life of the account rather than 28 days, per the same changelog. Code-review default effort moves from Lite to Balanced on the same date.
Two things this article will not do. It will not convert those figures into euros — they are US dollars, and quietly reprinting them with a euro sign is how a number stops being true. And it will not turn the credit drop into a headline percentage: the arithmetic is ours, not GitHub's, so the honest form is the two dollar figures side by side.
Why did my Copilot bill change when the seat price didn’t?
Because the seat price and the seat's contents are now two separate variables, and only one of them appears in your contract.
A per-seat licence used to be a complete description of what you were buying. You knew the unit cost, you multiplied by headcount, and the result was the annual commitment. Usage-based components break that. The seat price becomes an entry fee; the included allowance becomes a quota; anything past the quota becomes a variable line that arrives after the fact.
The practical consequence for a finance function is specific: the number you budgeted against has stopped being the number that decides your invoice. Your renewal review is pointed at the seat price, which is stable, and the movement is happening somewhere your review does not look.
What is per-seat-plus-consumption pricing?
It is a licence that charges a fixed fee for access plus a variable fee for use, with an allowance of use bundled into the fixed fee. Microsoft described the direction on the record in 2026 as moving "beyond per-seat to per-seat-plus-consumption." GitHub's schedule is the first instance we have seen documented by the vendor itself with dates attached.
The reason it matters more for AI tools than for the software that came before them is that AI usage is genuinely variable per person. Two people on identical seats can differ by an order of magnitude in what they consume, and neither of them can see the meter. Traditional licence management assumes the opposite — that a seat is a seat.
Does this affect Microsoft 365 Copilot?
No. GitHub Copilot and Microsoft 365 Copilot are different products on different bills, and conflating them in front of a CFO is a real error. GitHub Copilot is developer tooling; the figures above apply to it and to nothing else.
The *pattern* generalises. The *numbers* do not. If you want the M365 side, we wrote separately about the 2026 Microsoft Copilot and M365 price changes, which moved in the more familiar direction: the seat price itself went up.
The 1 October rule is the part that should bother you most
From October, a seat bills the moment an administrator assigns it. Not when the person signs in. Not when they use it.
Some IT teams have used the gap between assignment and activation deliberately — staging a rollout, pre-provisioning a department, holding seats for a cohort that starts training next month. That gap is now billable from day one.
Which is precisely the "paying for seats nobody opens" problem with the grace period removed. It does not create the waste; it accelerates the bill for waste that was already there. If a quarter of your assigned seats have never been opened, you were already paying for them — you will now simply pay sooner and with less warning.
There is also a governance detail nobody will be told about in a budget meeting: chat data retention goes from 28 days to indefinite by default from late September. That is a question for whoever owns your data-retention policy, and it is better asked before the change than after.
What to check before the October invoice
Five concrete things, in the order that produces the fastest answer:
- Count assigned seats against seats with activity in the last 30 days. Not licences purchased — seats *assigned*. The difference is what starts billing upfront in October.
- Find out who can assign a seat. If seat assignment is delegated to team leads and it now bills on assignment, you have moved a spending decision to people who do not see the invoice.
- Look at credit consumption per user, not in aggregate. An average tells you nothing here. You are looking for the shape of the distribution — how many people are nowhere near the allowance, and how many are through it.
- Ask what happens at the allowance ceiling. Does work stop, degrade, or continue and bill? Each of those is a different operational risk and a different conversation.
- Put the retention change in front of whoever owns retention policy before 28 September.
None of that requires a tool purchase. All of it requires somebody to own the question.
The underlying point, which outlives this changelog
A cost line you cannot see the inside of is not under management. It is under observation.
The reason this shift catches finance teams is not that vendors are being sneaky about it — GitHub published the schedule months in advance and again last week. It is that the internal control was built around the wrong variable. Renewal reviews check price. Metered licences change value. Those are different quantities and only one of them was ever being watched.
If you are tracking licence *cost* and not licence *usage*, you have no instrument that can see this coming — not because the information is hidden, but because you are not pointed at it. The fix is unglamorous: a real utilisation number per licence, produced independently of the vendor whose renewal is at stake, with a euro figure attached to the gap. That is week one of a waste audit, and it is the same work whether the trigger is a metering change or a renewal.
The seat price is what you agreed to. The seat's contents are what you actually bought. Only one of those is in the contract.