What Are GitHub Copilot Premium Requests?

What Are GitHub Copilot Premium Requests?

If you went looking for your premium request balance recently and could not find it, you are not going mad. The thing you are looking for was retired as the main way GitHub Copilot is billed, and for most accounts it no longer exists.

Premium requests were the metering unit GitHub used for Copilot from mid-2025. You got an allowance each month, heavier models drew down that allowance faster, and when you ran out you either paid for more or dropped to a cheaper model. It was simple to explain and simple to budget for, which is exactly why so many teams built their internal cost expectations around it.

On June 1, 2026 GitHub replaced that system. This guide covers what premium requests were, what took their place, who is still being billed on the old model, and the one group of users who need to do something about it before their plan expires.

What Are GitHub Copilot Premium Requests?

A premium request, or premium request unit, was a single billable interaction with one of Copilot's more capable models. Each interaction cost one unit, which was then multiplied by a model-specific number, so a request to an expensive model could consume several units at once.

As of June 1, 2026, GitHub's own documentation describes premium request billing as legacy. It applies to one group only: Copilot Pro and Pro+ subscribers on an existing annual plan who remained on request-based billing past that date. Everyone else is now billed on usage, measured in GitHub AI Credits.

So the honest answer to the question depends on what you are holding. If you are on a monthly plan, premium requests are history and the number you want to understand is your credit allowance. If you are on an annual Pro or Pro+ plan bought before the change, they are still live for you and will be until that plan runs out.

Not sure where your own tooling spend is going? Get a free SEO audit and start from data rather than invoices.

How Did Premium Requests Work?

Worth understanding even now, partly because annual subscribers are still on it and partly because the shape of the old system explains why GitHub changed it.

The Unit and Its Multipliers

One model interaction equalled one premium request, scaled by a multiplier attached to whichever model served it. Lighter models drew less, heavier ones drew more, and the multiplier was the only lever that connected cost to the actual expense of answering you.

That multiplier is also where the system quietly got more expensive over time. GitHub's documentation notes that from June 1, 2026 Copilot code review carries a model multiplier of 13, meaning one review consumed thirteen units of allowance. Annual subscribers who stayed on legacy billing saw their multipliers increase on that date even though their plan price did not change.

The Monthly Allowances

Under request-based billing the published allowances were:

• Copilot Pro at 10 US dollars per month: 300 premium requests
• Copilot Pro+ at 39 US dollars per month: 1,500 premium requests

Counters reset on the first of each month at 00:00:00 UTC, and unused requests did not carry over. A quiet month bought you nothing the following month, which is a detail that annoyed plenty of people who worked in bursts.

Running Out and Paying for More

Additional premium requests cost 0.04 US dollars each. GitHub notes these were not available to users who subscribed through GitHub Mobile on iOS or Android, which caught out a small but genuinely confused group of users whose purchase route determined whether they could buy an overage at all.

The other thing that happened when you ran out was automatic: Copilot fell back to a lower-cost model and kept working. You lost capability rather than access. That fallback is one of the behaviours that has now ended, and it matters more than it sounds.

A Short Timeline

For anyone reconciling old invoices, the dates GitHub documents are:

• June 18, 2025: premium request billing begins for paid plans on GitHub.com
• August 1, 2025: the same begins on GHE.com
• November 1, 2025: premium requests for Spark and the cloud agent move to dedicated SKUs
• June 1, 2026: usage-based billing replaces premium requests for all plans except existing annual ones

Want the same clarity about your search performance? Run a free SEO audit and see what is actually driving your traffic.

What Replaced Premium Requests?

GitHub AI Credits, under a usage-based billing model that started June 1, 2026. Instead of counting interactions, Copilot now counts what each interaction consumes.

Credits Are Measured in Tokens, Stated in Dollars

Credits are drawn down by token use, covering input, output and cached tokens, charged at each model's published API rates. The included allowance in every plan is expressed as a dollar figure rather than a request count:

• Copilot Pro: 10 US dollars in AI Credits per month
• Copilot Pro+: 39 US dollars in AI Credits per month
• Copilot Business: 19 US dollars per user per month
• Copilot Enterprise: 39 US dollars per user per month

GitHub also published a transitional uplift for existing customers: Business accounts received 30 US dollars per month and Enterprise accounts 70 US dollars per month for June through August. Base plan prices did not change at the switch.

The practical difference is that cost now tracks session length. Under the old model a one-line question and a two-hour agentic session both cost one request times a multiplier. Under the new one, the long session costs what it consumes, which is a lot more.

What Still Does Not Cost You Anything

Code completions and Next Edit suggestions remain included and do not consume credits. This is the single most reassuring detail in the change for most developers, because inline completion is the feature people use constantly without thinking about it. The metering applies to chat, agents and the heavier model work.

One exception to watch: Copilot code review also consumes GitHub Actions minutes, billed at the standard per-minute rate, so its true cost appears in two places on your bill rather than one.

Pooling and Budgets for Teams

For Business and Enterprise, included usage is pooled across the organisation rather than ring-fenced per seat. One heavy user can draw on the pool, which is either a feature or a problem depending on your team.

To control that, admins can set budgets at the enterprise, cost centre and individual user level. When the pooled allowance runs out, the organisation decides whether to permit further usage at published rates or to cap spending there. Given that the old automatic fallback to a cheaper model is gone, setting those budgets deliberately is now the thing standing between a busy sprint and a surprising invoice.

Who Still Has Premium Requests, and What Should They Do?

Only Copilot Pro and Pro+ subscribers on an annual plan who stayed on request-based billing after June 1, 2026. GitHub documents three options for them.

Option 1 - Stay on the Annual Plan

Staying put keeps the plan tracking usage in premium requests with model multipliers, at the price already paid. When the annual term ends, the account moves to Copilot Free rather than rolling into a paid monthly plan, so this is the option with a cliff at the end of it. Put the expiry date in a calendar now, because the day it lapses your team loses paid Copilot without anyone making a decision.

Option 2 - Cancel for a Prorated Refund

Cancelling returns a prorated refund for the unused portion of the annual term, with the option to resubscribe to the equivalent monthly plan afterwards. This is the cleanest route if you want to reset and reassess rather than carry a legacy arrangement you have to keep explaining to whoever approves the spend.

Option 3 - Upgrade to a Monthly Plan

Upgrading moves you to a monthly paid plan immediately and gives prorated credit for the remaining annual value, so the money already spent is not lost. This is the option to take if your usage has shifted towards agents, since the increased multipliers make the old allowance drain faster than it did when you bought the plan.

Which is better depends entirely on how you work. If your usage is light and steady, the old allowance may still be the better deal for the months remaining. If you have started using agents for long multi-step tasks, the multipliers that increased on June 1 will chew through 300 or 1,500 units faster than they used to, and moving across early is worth modelling.

Getting your budget decisions right starts with knowing what works. Start with a free SEO audit and prioritise from evidence.

Why Did GitHub Change the Model?

GitHub's stated reasoning is that Copilot stopped being a completion tool and became an agentic platform that runs long, multi-step sessions, and those sessions need far more compute than a chat message. Under request-based billing a quick question and a multi-hour agent run cost the same, which meant the price signal had stopped describing the product.

The company said it had absorbed much of the rising inference cost and that the premium request model was "no longer sustainable". It also framed usage-based billing as reducing the need to throttle heavy users, which is the flip side of removing the automatic downgrade.

Whether that reads as fair or as a price rise depends mostly on how you use it. Light chat users on Pro are arguably better off, since a dollar allowance spent on cheap models goes further than 300 requests did. Heavy agent users are the ones with a genuine new cost exposure, and they are also the ones who most liked the old flat ceiling.

Premium Requests vs AI Credits

The two models differ in four ways worth holding onto.

• What gets counted: interactions scaled by a model multiplier, versus tokens consumed at published model rates
• What the allowance looks like: a request count, versus a dollar figure
• What happens at the limit: an automatic fall back to a cheaper model, versus a budget decision made by you or your admin
• How predictable the bill is: highly predictable but poorly matched to real cost, versus closely matched to cost and considerably harder to forecast

That last trade is the whole change in one line. You have swapped a ceiling for a meter. The work that used to be done by GitHub's throttle is now done by your budget settings, and if you never set them, nothing is doing it.

If the arithmetic stops working for you, our walkthrough of how to cancel a GitHub Copilot subscription covers the steps and the billing-cycle timing that decides whether you pay for one more month.

Final Thoughts

Premium requests are worth understanding as history and, for annual subscribers, as a live bill with an expiry date attached. For everyone else the number that matters is the credit allowance, and the habit that matters is checking consumption during the first heavy week of agent use rather than at the end of the month.

Two things are worth doing this week whichever plan you are on. Find your current usage page and look at what a typical day actually consumes, because almost nobody's intuition about token cost is correct. And if you administer a team, set the budgets, since the safety net that used to catch overruns automatically has been removed by design.

Prices, allowances and multipliers in this area have moved twice inside a year. Treat every figure here as correct as of the dates given and check GitHub's own pricing and billing pages before committing a team to a plan.

Ready to put the same rigour into your search strategy? Get a free SEO audit and see where the real opportunities are.