September 1, 2026 ·

github-copilotcodingpricingprocuremententerprisebillingagents

GitHub Copilot's price did not change today. Your included budget fell by up to 44%.

TL;DR: The promotional AI credit allowance GitHub granted existing Copilot Business and Enterprise customers for June, July and August 2026 ends today, 1 September. Per GitHub’s own billing documentation, included usage reverts to the standard amounts: Business from 3,000 to 1,900 credits per user per month, Enterprise from 7,000 to 3,900. At the published rate of 1 credit = $0.01, that is a fall from $30 to $19 and from $70 to $39 of included consumption per seat — reductions of 37% and 44%. Seat prices do not change: $19 and $39 per user, exactly as before. Code completions and next edit suggestions remain free and unlimited — credits are consumed by chat, agent mode and premium model requests. The thesis: this is the point at which June’s move to usage-based billing actually starts costing existing customers money, and it arrives with no price rise, no renewal and no invoice line that moves. For you: compare your last three months of per-seat consumption against 1,900 / 3,900, not against what you have been watching, and check whether your org is set to overage or to a hard cap.

What changes today

On 1 June 2026 GitHub moved Copilot to usage-based billing. The headline structure was a one-to-one ratio of plan price to monthly entitlement: a $10 plan includes $10 of AI credits, a $39 plan includes $39. Tokens consumed across chat, agent mode and premium model requests draw down that allowance at published per-model rates. Code completions and next edit suggestions stayed outside the meter entirely, unlimited on every paid plan.

What was less visible, and is the reason today matters, is that existing Business and Enterprise customers were not put on that ratio immediately. GitHub gave them an elevated allowance for the first three months of the new regime — 3,000 credits per user per month on Business and 7,000 on Enterprise, against standard amounts of 1,900 and 3,900. The promotional window covered June, July and August. It closes today, and included usage returns to the standard tiers.

The arithmetic is straightforward at one cent per credit:

Plan	Seat price	Included credits (Jun–Aug)	Included credits (from 1 Sep)	Change
Copilot Business	$19/user	3,000 (~$30)	1,900 (~$19)	−37%
Copilot Enterprise	$39/user	7,000 (~$70)	3,900 (~$39)	−44%
Copilot Pro	$10	$10	$10	unchanged
Copilot Pro+	$39	$39	$39	unchanged

Credits pool at the organisation level rather than being locked to individual seats, and they do not roll over — the allowance resets on the first of each month.

Why this will catch people

The uncomfortable part is not the size of the reduction. It is the shape of it.

A price increase is a legible event. It produces a notification, a renewal discussion, a procurement review, and in most organisations an approval step where somebody has to look at the number and agree to it. None of that happens here. The seat price on the September invoice is identical to the August one. There is no contract change, no renewal date, and nothing in the billing line that a finance team scanning for variances would flag.

The change is visible in exactly one place — the ratio between what a team consumes and what it is allotted — and that ratio lives on a dashboard nobody is obliged to open.

Worse, the population most exposed is the population that behaved most responsibly. When usage-based billing landed in June, the sensible advice was to watch the credit dashboard for a month before drawing conclusions, and GitHub shipped the tooling to make that possible: per-user credit consumption in the usage metrics API from 19 June, per-billing-cycle credit visibility in the product from 20 July, AI credit pools in cost centers from 2 July. Teams that took that advice spent June, July and August measuring real consumption against a 3,000 or 7,000-credit ceiling, found themselves comfortably inside it, and closed the ticket.

Every one of those measurements was taken against a cushion that does not exist from today. A Business team averaging 2,400 credits per seat looked like it had 20% of headroom to spare all summer. From this morning it is 26% over the line.

Who is actually exposed

Not everyone, and the distribution matters more than the average.

The exemption of code completions is doing enormous work here. Tab-completion and next edit suggestions — the original Copilot product, and still the way most developers touch it most of the time — consume no credits at all and remain unlimited. A team whose usage is predominantly inline completion will not notice today happened, and no amount of coverage should convince them otherwise.

Credits are consumed by the conversational and autonomous surfaces: chat, agent mode, and premium model requests. That means exposure concentrates in a minority of developers running heavy agentic workflows against large repositories — the same pattern that produced reports of 10-50× bill increases when the metered model first landed, and the same pattern GitHub and every competitor have spent a year encouraging everyone to adopt.

So the practical question is not “what is our average consumption.” It is “which seats are consistently above 1,900 or 3,900, and what do we want to happen when they hit it.” In most organisations that is a short list of names, and treating it as an org-wide problem will produce a worse decision than treating it as a per-seat one.

The pattern this belongs to

Strip out the specifics and this is the third instance of the same mechanism in as many weeks, which is what makes it worth more than a billing note.

AI vendors have converged on introductory pricing with a documented expiry date. Sometimes the clock runs out and the price rises, as with OpenAI’s promotional rate expiry. Sometimes the vendor cancels the increase and the introductory rate becomes permanent, as Anthropic did on 11 August when it scrapped the scheduled rise on Claude Sonnet 5 and made $2/$10 the standard price. And sometimes, as today, the price never moves at all and the entitlement behind it quietly reverts.

The through-line is that the cost of an AI subscription in 2026 is increasingly set by an expiry schedule rather than by a rate card. A price you were quoted six months ago tells you progressively less about what you will pay. The only durable defence is unglamorous: keep a dated list of every promotional period, introductory rate and elevated allowance across your AI vendors, with its end date, and review it monthly. Most organisations track renewal dates and track nothing else, which is precisely why a change like this one lands unseen.

What to do

Four steps, well under an hour of work.

Pull three months of per-seat consumption. The Copilot usage dashboard and the usage metrics API have both reported per-user credit consumption since June, so the data exists whether or not anyone has looked at it.

Re-baseline against the new numbers. Compare each month against 1,900 credits per user on Business or 3,900 on Enterprise. The promotional figures you have been watching are no longer the relevant line.

Check your overage setting. Organisations can either allow overage billing at published per-model rates or cap spending at the included allowance. Overage means work continues and the invoice grows; a cap means the invoice is predictable and premium requests stop until the monthly reset. Both are defensible; not knowing which one is configured is not. Cost centers support AI credit pools, so caps can be scoped to teams rather than applied bluntly to everyone.

Decide per seat, not per org. For the small group consistently above the line, the choice is a deliberate one: fund the overage, cap them, move them to a different tool for agentic work, or restructure the workflow. If that group turns out to be a large share of your engineers, run a proper total-cost comparison — including realistic overage — against Cursor, Claude Code and OpenAI Codex rather than comparing seat prices, which at this point is comparing the least informative number on offer.

GitHub Copilot moves to token billing — the June transition this completes
GitHub Copilot review · Cursor review · Claude Code review
GitHub Copilot vs Cursor · Cursor vs Claude Code
Best AI coding tools · Best AI harnesses

Frequently asked questions

Am I affected if I'm on Copilot Pro or Pro+?

No. The promotional allowance applied to existing Copilot Business and Copilot Enterprise customers only, and individual plans were never part of it. Copilot Pro remains $10 per month with $10 of included credits and Pro+ remains $39 with $39 of included credits — the same one-to-one ratio of plan price to monthly entitlement that has applied since 1 June. If you are an individual subscriber, nothing about your bill changes today. The teams affected are organisations that were already on Business or Enterprise seats before the June transition, which is the population GitHub was cushioning while everyone learned what agentic workflows actually consume.

Do code completions eat into the reduced allowance?

No, and this is the single most important thing to understand before panicking about the reduction. Code completions and next edit suggestions are not billed in AI credits and remain unlimited on all paid plans. Credits are consumed by chat, agent mode, and premium model requests — the conversational and autonomous surfaces, not the inline tab-completion that most developers use most of the time. A team whose usage is predominantly tab-complete will very likely never approach either the old allowance or the new one, and for them today is a non-event. The exposure is concentrated entirely among people running agent mode against large repositories, which is also the usage pattern GitHub has spent a year encouraging.

What happens when a team exhausts its credits now?

It depends on a setting your administrator controls, and checking which way it is configured is the highest-value thing to do this week. Credits are pooled at the organisation level rather than locked per user, and unused credits do not roll over — they reset on the first of each month. If overage billing is enabled, work continues past the allowance and is charged at published per-model rates, which is convenient and is how surprise invoices happen. If spending is capped at the included allowance, the bill is predictable and premium requests stop until the reset, which is safe and is how a sprint stalls mid-week. Neither is wrong. What is wrong is not knowing which one you have, because you will find out in the least convenient way.

Is Copilot still competitive after this?

On seat price it remains among the cheapest ways to put a coding assistant in front of a large team, and that has not changed today — $19 and $39 per user are unchanged. What has changed is the honesty of the comparison. Copilot Business at $19 with $19 of included consumption is a different product from Copilot Business at $19 with $30 of included consumption, and for the past three months the second one is what existing customers actually had. Compared against Claude Code or Cursor subscriptions, the right way to run the numbers is total monthly cost per developer including realistic overage, not the seat price. For heavy agentic users that comparison has moved, and for light users it has not moved at all. Which group you are in is an empirical question your credit dashboard can answer in about five minutes.

Why does this land as a reduction rather than a price increase?

Because structurally it is the expiry of a discount, not a repricing, and that distinction has real consequences for how organisations catch it. A price increase generates a notification, a renewal conversation, a procurement review and usually an approval step. A promotional allowance reverting to its documented standard generates none of those. The subscription line on the invoice is identical in September to what it was in August. The change surfaces only as a shift in the ratio between consumption and allowance, which is visible on a dashboard nobody is required to look at. This is a general pattern worth internalising for AI tools specifically, because introductory pricing with a documented end date has become the industry norm — the cost of an AI subscription is increasingly set by an expiry schedule rather than by a rate card.

What should we actually do this week?

Four things, in order, and the whole exercise is under an hour. Pull the last three months of credit consumption per seat from the Copilot usage dashboard or the usage metrics API, which has reported per-user credit consumption since June. Compare the monthly figures against 1,900 credits for Business or 3,900 for Enterprise rather than against the promotional numbers you have been watching. Identify the seats that were consistently above the new line — in most organisations this is a small minority of developers running heavy agent workloads, not an even distribution. Then make a deliberate decision on overage versus a hard cap for those seats specifically, rather than letting the default apply to everyone. If a meaningful share of your team sits above the line, that is the signal to price a comparison against alternatives properly.

Sources

Related tool reviews

Questions or corrections? Email Pick Right. Want the full list? See all news.