GitHub is replacing premium requests with AI Credits, which makes long-running Copilot sessions look less like a flat subscription perk and more like a capacity policy teams will have to manage.
GitHub just made something explicit that the coding-tool market has been avoiding for a while.
A quick inline suggestion and a long-running agentic coding session are not the same product.
They do not cost the platform the same thing. They do not create the same budget risk. And they should not be governed the same way.
That is why GitHub's move from premium requests to AI Credits matters.
On paper, this is a billing transition scheduled for June 1. In practice, it is GitHub admitting that Copilot is no longer just an assistant feature. It is becoming a compute-bearing workflow surface that teams will have to budget and operate more deliberately.
GitHub says all Copilot plans are moving to usage-based billing on June 1, 2026.
Premium request units are going away. GitHub AI Credits are taking their place.
Usage will be calculated from token consumption, including input, output, and cached tokens. GitHub also says base plan prices are not changing, while additional consumption can draw down credits and trigger paid overage-style usage for qualifying plans.
That is the factual part.
The more interesting part is GitHub's explanation for why the change is happening.
The company says Copilot has evolved into an agentic platform capable of longer, multi-step sessions across larger repositories, and that the old model no longer matched real inference cost.
That is the real story.
Premium requests were easier to understand, but they hid too much.
A lightweight interaction and a heavy autonomous session could feel similar from the user side even when they were very different from the platform side. That mismatch works for a while when the product is mostly chat and completion assistance. It breaks once users start expecting multi-step coding help that can run longer, inspect more context, and burn far more compute.
AI Credits are GitHub's way of making that underlying cost model more visible.
That visibility will probably frustrate some users. It may also lead to better operating habits.
Once usage is visibly metered, teams have stronger reasons to ask which tasks deserve the premium path, which models deserve default routing, and where cheaper fallback behavior is good enough.
The strongest Butler angle here is administrative, not emotional.
If Copilot becomes a more explicit credit consumer, then admins need visibility, budgeting expectations, and clear rules for heavy-use workflows. GitHub's preview bill experience points in that direction too. The company clearly expects buyers to start monitoring spend before the new model becomes live.
That is a sign of maturity, but it is also a sign that agentic coding is graduating into a policy problem.
The closer Copilot gets to running longer sessions with frontier models across bigger codebases, the less credible it becomes to govern everything with the same subscription-era mental model.
Teams do not need to panic, but they do need to get specific.
Figure out where Copilot is being used for:
Those behaviors should not all inherit the same budget assumptions.
This is also a good moment to revisit the kind of model-routing discipline we talked about in how GitHub already turned cloud-agent usage into a routing question. Once heavy usage has a more explicit meter, routing and escalation policy stop being nice-to-have optimizations and start looking like normal operational hygiene.
GitHub is not alone here.
Across the market, coding tools are being forced to reprice around actual autonomy, actual context size, and actual session duration. The more agentic the workflow becomes, the harder it is to preserve the illusion that everything should feel flat-rate.
So this is not just a GitHub story.
It is another sign that AI coding is entering its capacity-planning era.
Teams that treat it that way will probably make better decisions than teams that keep pretending every AI interaction is basically the same unit of work.
This article was researched and drafted with AI assistance, then reviewed and edited for clarity, accuracy, and editorial quality.