GitHub Copilot's Meter Is Now Running: Usage-Based Billing Arrives June 1
GitHub just told you what Copilot actually costs to run. That matters more than the price change.
Starting June 1, Copilot stops being a subscription and starts being a meter. GitHub is replacing its premium request unit model with GitHub AI Credits on June 1, which means every chat prompt, every CLI session, every code review pass, and every cloud agent turn will burn tokens at published API rates. The flat monthly fee stays on the label, but the effective cost for heavy users is no longer capped. If you run a multi-hour autonomous session, you will pay for it.
The announcement came with the usual vendor framing — more flexibility, more visibility, more control. And underneath that, a very honest number: Copilot's week-over-week costs nearly doubled between January and now, driven largely by agentic workflows that consume tokens at rates the original flat-rate model never anticipated. That is not a pricing grievance. That is a product architecture confession. GitHub built Copilot around the assumption that AI assistance would mostly mean autocomplete and chat. Then users started running agents that never stopped, and the math broke.
The mechanics are straightforward. Pro and Pro+ get $10 and $39 in monthly credits respectively. Business and Enterprise get $19 and $39 per user per month, with a promotional bump to $30 and $70 through August. Code completions and Next Edit suggestions stay included — no credits consumed. But everything else — chat, CLI, code review, cloud agents — runs against the meter. And critically, when credits run out, usage stops. There is no fallback to a lower tier, no graceful degradation, no "we will let you know when you hit the wall." The product just stops.
That last part is the part that should make engineering leads pay attention. In the current model, hitting your request limit is annoying. In the new model, it is a workflow interrupt at the worst possible moment — mid-session, mid-debug, mid-PR review. GitHub removed the fallback experience that gave users *some* continued access when limits hit. The stated reason is simplicity. The actual effect is that Copilot is no longer reliable as background infrastructure if your usage is uneven. You either have enough credits for a full workday or you have a broken tool.
The code review twist is particularly interesting. Starting June 1, Copilot code review will also consume GitHub Actions minutes at standard runner rates. That is a meaningful conflation. Code review is now partly a compute billing event, which means the feature has to justify itself on two dimensions — does it catch real bugs, and does the compute cost less than the reviewer time it replaces? Teams that enabled code review as a safety net may suddenly find themselves doing that math for the first time.
For practitioners, the immediate action is to wait for the early-May preview bill tool and actually use it. Not to estimate, to model. Run your team's Copilot sessions through the forecast tool before June 1 and see what the realistic burn rate looks like. If your CLI users run 4-hour autonomous sessions, or if your PR volume drives heavy review usage, the $39 Pro+ allocation may not survive a full workday. Business and Enterprise have more runway with the promotional credits, but those disappear after August. The question to answer before September is not "is Copilot worth it?" It is "which workflows am I willing to meter?"
The bigger picture is that this is the second major AI coding tool in weeks to explicitly move away from flat-rate pricing. OpenAI shifted Codex to token-based team pricing in early April. GitHub is following the same direction with Copilot. The difference is that Copilot sits inside GitHub's platform, so the billing gets tangled with Actions minutes and enterprise commit logs. That is actually important: Copilot is not just a coding assistant anymore. It is a platform component whose costs now flow through the same budget line as your CI/CD pipeline. CFOs and engineering leads who have been treating AI coding tools as a line-item SaaS decision are about to have a very different conversation.
What GitHub is really doing here is drawing a line under the subsidized era. AI coding tools got cheap flat rates because vendors were subsidizing adoption, burning through VC money, and betting that usage would stabilize below the flat fee's cost. It did not. Agentic workflows changed the unit economics faster than anyone projected. The response is not a price hike — it is a structural repricing that makes the actual cost visible. That is honest, in its way. It is also going to force every team that deployed Copilot loosely to start thinking about it like infrastructure: monitored, budgeted, and deliberately scoped.
The developers who will feel this most are the ones who have been treating Copilot like an always-on background process — agents running in terminals, automated PR reviews, long coding sessions that chew through context. For them, June 1 is a discontinuity. Everyone else will feel it as a subtle shift in behavior: more awareness of when the meter is running, more discipline about what deserves Copilot's attention and what does not. That is not necessarily bad. Metered tools often get used more deliberately. But it is a different relationship with the product than "pay $39 and forget it."
GitHub will tell you this gives you more control. They are not wrong. Admins get budget caps at enterprise, cost center, and user levels. Usage is visible per-feature and per-model. You can actually see where the money goes now. That is better than a black box. But control only matters if you use it — and the teams that will thrive in this model are the ones that start treating Copilot like a metered cloud service rather than a flat-rate subscription. The ones who will be surprised are the ones who do not.
Sources: GitHub Blog, Ars Technica