A developer opens their billing dashboard on a Tuesday morning, not because anything went wrong, but out of habit. One agentic session the day before - "fix the bug in the auth flow" - has consumed 1,180 credits. That is 16% of a monthly Pro+ allowance. The model gave mediocre suggestions. They did most of the work themselves.
That specific story, reported by The Register, is not an edge case. It is the point.
On June 1, 2026, GitHub moved all Copilot plans to usage-based billing. Instead of counting premium requests, every plan now includes a monthly allotment of GitHub AI Credits, with the option to purchase more.
Usage is calculated based on token consumption - input, output, and cached tokens - using the listed API rates for each model.
Base subscription prices stayed the same on paper. Copilot Pro remains $10 per month, Pro+ remains $39. But those figures now describe included credit value, not the practical ceiling of what a power user might consume.
The distinction matters enormously. The backlash was immediate because the new meter exposed a truth vendors had been trying to smooth over: AI coding is not priced like software, it is priced like compute.
Why agentic sessions are expensive
The old Copilot interaction was small: you typed a line, it autocompleted. The new interaction is large: you describe a task, the model inspects multiple files, constructs context, generates patches, re-evaluates, and iterates. When a user asks an AI assistant to "fix this bug," the assistant may inspect multiple files, construct context, generate patches, re-evaluate errors, and produce output much larger than the original prompt. The user experiences one request. The system experiences a chain of token-consuming operations.
Copilot has evolved from an in-editor assistant into an agentic platform capable of running long, multi-step coding sessions. Agentic usage is becoming the default, and it brings significantly higher compute and inference demands. Today, a quick chat question and a multi-hour autonomous coding session can cost the user the same amount - which it no longer will.
GitHub's reasoning is credible. GitHub absorbed much of the escalating inference cost behind that usage, but the current premium request model was no longer sustainable. That is fair. What is less fair is that the product actively encouraged the behavior now being re-priced.
The whole pitch of AI coding assistants is that they make iteration cheaper, faster, and less intimidating. If GitHub did not want users to lean into agentic workflows, it would not have built so much product around them.
What developers are actually reporting
Developers on GitHub's discussion thread report heavy backlash, with more than 400 comments and nearly 900 downvotes; some users report burning substantial portions of a month's credits in a single session.
One developer on the $39 Copilot Pro+ plan used about 8% of their monthly AI credit allotment in just two hours, estimating that their 7,000-unit quota might run out in less than two days.
One developer estimated their company's bill would jump from $29/month to $750/month. Another projected cost rose from $50 to $3,000.
Some of those projections reflect genuinely wasteful habits - asking the model to rewrite large chunks of a project without reading the output, iterating blindly. Token-based billing exposes that waste in a way flat pricing concealed. Fair enough. But the counterargument that only "vibe coders" will see big bills misses something real: agentic coding is inherently unpredictable in token terms. A legitimate, careful developer asking a model to reason across a complex codebase will consume a lot of tokens. That is not sloppiness - it is the use case.
There is also a new wrinkle for teams using code review
Copilot code review moved to an agentic architecture that runs on GitHub Actions, and starting June 1, reviewing a pull request with Copilot counts against included Actions minutes at the same per-minute rates as any other Actions workflow.
For teams that rely heavily on automated code review, this could add unexpected costs on top of the new credit system.
A team that already runs heavy CI pipelines may now find Copilot review nudging them into Actions overages they did not budget for. That is worth a quick audit before the next billing cycle closes.
What to actually do this week
Before your next sprint planning, three things are worth doing:
- Check April and May usage reports. GitHub surfaced a preview billing tool in early May; those historical numbers will show what your team's real consumption pattern looks like under the new model.
- Set a hard budget cap at the org level. Administrators can set spending limits in the GitHub Copilot billing dashboard to control monthly charges. Set one before someone else discovers the meter the hard way.
- Audit which workflows use frontier models by default. Choosing less expensive models for routine tasks can help, with higher-cost models reserved for more complex work. Most inline completions do not need the heaviest model available. An hour with the model settings is worth more than a month of reactive cost management.
The bigger shift underneath the billing
The controversy around Copilot's new billing is less a freak-out than a market correction. AI coding assistants are leaving the promotional phase and entering the managed-infrastructure phase, where the winners will be judged not only by how much code they can produce, but by how predictably, governably, and economically they can help humans ship software.
That reframe applies to every AI tool your team uses, not just Copilot. The subsidized exploration phase - where vendors absorbed inference costs to build habits - is ending. For organizations budgeting developer tooling, the practical outcome is less predictability: the same monthly subscription now represents both a fixed access tier and a finite usage pool.
Treat your AI tool budget the same way you treat your cloud bill. Set limits. Watch the meter. Choose models deliberately. Teams that do that now will not be surprised by the next billing change. And there will be a next one - because every provider running frontier models at scale is having the same internal conversation GitHub just had in public.