Scale · founder · 7 min read

Your Copilot Bill Goes Up on Tuesday. The Price Doesn't Change.

On September 1, GitHub Copilot's promotional AI credits expire. Business drops 37%, Enterprise 44%, and the subscription price stays exactly the same.

Updated August 31, 2026 ·

pricing github copilot budget agents anthropic

On Tuesday, September 1, GitHub Copilot’s promotional AI credits expire.

Copilot Business drops from 3,000 included credits per user per month to 1,900. Copilot Enterprise drops from 7,000 to 3,900. That’s a 37% cut and a 44% cut.

Copilot Business is still $19 per user. Copilot Enterprise is still $39. Nothing on the pricing page moves.

This is the clearest example yet of something this site has been circling all year: for AI coding tools, the sticker price has stopped being the price. The number that determines your bill is the included allowance, and that number changes without a press release.

What actually expires

When Copilot moved to usage-based billing on June 1, existing customers got a cushion. GitHub’s own billing docs describe it plainly under the heading “Promotional amounts for existing customers”:

Existing Copilot Business and Copilot Enterprise customers receive a higher amount of included AI credits for the first three months of usage-based billing (June 1 – September 1, 2026).

Three months. June, July, August. Then, in GitHub’s words, “included usage returns to the standard amounts above” — 1,900 for Business, 3,900 for Enterprise.

A credit is worth exactly one cent. GitHub fixes it: 1 AI credit = $0.01 USD. So the arithmetic is unusually easy for once:

• Business: you lose 1,100 credits per user per month. That’s $11 per seat of included usage, gone.
• Enterprise: you lose 3,100 credits per user per month. That’s $31 per seat, gone.

On Business, you’re losing $11 of allowance on a $19 seat. The subscription didn’t get more expensive. It just started covering a lot less.

The part that will actually surprise people

Credits are pooled across your billing entity, not handed out per person. Twenty Business seats means one shared pool of 38,000 credits, not twenty buckets of 1,900. That’s genuinely good design — heavy users draw more, light users subsidize them. It also means nothing warns an individual that the team is running out.

Then there’s the default. From the docs:

Additional usage is enabled by default for organizations and enterprises.

When the pool empties, usage doesn’t stop. It keeps going at published per-credit rates and bills to your org. To stop it, an admin has to go into AI Controls and explicitly disable the AI credits paid usage policy. There’s also no automatic fallback to a cheaper model when a budget runs dry — you either keep paying frontier rates or get cut off.

Put those facts together and September 1 looks different. Your allowance drops by more than a third, your team’s usage almost certainly doesn’t, and the overage tap is already open.

Here’s a 20-seat Business org whose usage stays flat at 60,000 credits a month:

	August	September
Subscription	$380	$380
Included credits	60,000	38,000
Overage	$0	$220
Total	$380	$600

A 58% increase in what you pay, with no change in what you use and no change in the price. The first sign most teams get will be the October invoice.

One piece of good news, stated clearly: code completions and next edit suggestions are not billed in credits and remain unlimited on all paid plans. If you use Copilot as smart autocomplete, none of this touches you. Credits burn on chat, CLI, Copilot Spaces, Spark, the cloud agent, and third-party coding agents. Agentic use is the cost driver, as it has been at every tool this year.

Anthropic did the opposite thing on the same date

September 1 was also supposed to be the day Claude Sonnet 5’s introductory rate expired, taking it from $2/$10 per million tokens to $3/$15 — a 50% rise.

On August 10, Anthropic cancelled it. From the company’s own account: “We’re making Claude Sonnet 5’s introductory pricing permanent… and that price will remain unchanged.” The pricing page now lists $2/$10 with no end date.

So two of the biggest vendors in this stack had a promotional period ending on the same day. One let it lapse quietly. One made the discount permanent and got a news cycle out of it. Neither is charity or malice — both are competitive positioning at the layer where the money is now actually decided, which is no longer the seat price.

The sticker price stopped being the price

Look at the last two weeks across three vendors:

• Cursor (August 24) raised included usage and simultaneously moved Auto from a flat $1.25/$6 rate to the list price of whichever model your request gets routed to. Still $20 a month.
• GitHub (September 1) cuts included credits 37–44%. Still $19 and $39.
• Anthropic (August 10) cancelled a scheduled 50% increase. Rate unchanged.

Three moves, three directions, and in all three the advertised price is identical before and after. Every one of them changed your bill anyway, by moving the allowance, the metering unit, or the overage default instead.

This is what consumption pricing looks like once it matures. The seat fee becomes a cover charge, and the real pricing lever moves to a table in the billing docs that nobody has a calendar reminder for.

The practical consequence: comparing AI tools on monthly price is now close to useless. A $19 seat with 1,900 credits and overage on by default is not cheaper than a $39 seat with 3,900 credits and a hard stop. You can’t know which is cheaper without knowing your own consumption.

What to do before Tuesday

Pull your August credit usage. You need one number: credits your org consumed in August. Compare it to the new pool (seats × 1,900 for Business, × 3,900 for Enterprise). If August exceeds it, that gap in cents is your new monthly overage, starting now.

Decide about the overage tap deliberately. Leaving it on is a legitimate choice — hitting a wall mid-sprint is expensive too. Making that choice by accident isn’t. Set an enterprise spending limit even if you keep it enabled.

Set user-level budgets before you need them. They cap what one person can draw from the shared pool; without them, one enthusiastic agent user can drain a team’s month. A $0 budget blocks that user immediately, so don’t use it as a placeholder.

Diarize the reset. Credits don’t roll over. The pool resets at 00:00:00 UTC on the first of each calendar month regardless of your billing date, and unused credits are forfeited.

Bottom line

Nobody is doing anything underhanded here. GitHub documented the promotional window in advance and published the standard amounts right alongside it. It’s only stealthy in the sense that a cushion ending is much quieter than a price rise — and lands on your invoice exactly the same way.

The habit worth building isn’t outrage, it’s a recurring calendar entry. Once a quarter, open the billing page of every AI tool you pay for and compare your actual consumption against your current allowance. That’s a finance task now, not an admin chore, and on current evidence it will stay one.

We’ve started keeping that calendar ourselves: every dated AI coding tool pricing change we can verify, including the Cursor Enterprise deadline a week after this one.

For the wider pattern, see the June 2026 billing reset, which covered the switch that created this cushion, and the Cursor Auto repricing from last week.

Sources

• GitHub Docs — Usage-based billing for organizations and enterprises (promotional window, standard credit amounts, credit value, overage defaults, budget controls)
• GitHub Blog — Copilot is moving to usage-based billing
• Anthropic — Claude Platform pricing
• Anthropic on X — Sonnet 5 introductory pricing made permanent