GitHub moves Copilot to usage-based billing on June 1 and changes access rules

GitHub is changing the basic economics of Copilot. From June 1, the service will switch to pay-as-you-go billing, and for many teams that matters more than the price list itself.

Copilot used to be seen as a nearly unlimited assistant bundled into a subscription. Now it gets a hard credit counter. When the limit runs out, users can no longer use the service until access is topped up.

According to the description of the new approach, GitHub plans to show a preview version of the billing in early May. That means users and companies will see in advance how the limit gets consumed and which actions affect the bill most.

The move itself is not surprising. Copilot long ago moved beyond simple autocompletion to heavier scenarios, in which the model writes large chunks of code, explains changes, helps with refactoring and increasingly works as an agent rather than an ordinary editor suggestion.

For GitHub, this is a logical step given its cost structure. The deeper an AI tool is embedded in the development process, the more requests it handles and the heavier the load on infrastructure and models. A fixed subscription is convenient for sales, but it fits poorly with a product where some users make a few simple requests a day while others run complex tasks through Copilot for hours. Pay-as-you-go ties revenue to actual load and ends the subsidizing of the most active users at the expense of everyone else.

Procurement changes for enterprise customers too. Copilot used to be booked as a straightforward per-employee subscription. Now finance and IT leaders will want to see limits, spending forecasts and overage rules. This brings AI tools closer to cloud infrastructure: they are used not "in general" but within budgets, quotas and internal policies. The better GitHub explains this mechanism, the easier it will be to scale Copilot inside large organizations.

For developers and team leads, this also changes behavior within the team. When a tool has a limit, people use it more deliberately: they save Copilot for hard spots, lengthy explanations, test generation and routine refactoring, rather than for every short autocompletion. Interest in transparency is growing alongside: which actions cost more, how fast credits burn, whether budgets can be set per team, and where the line runs between "included in the subscription" and "requires additional payment." The answers to these questions will shape the market's reaction more than the switch itself.

Another important point is the loss of the sense of unconditional availability. Under the new scheme, running out of credits means not lower quality and not slower responses, but an actual stop of the service. For some users this will be an annoying restriction, especially if Copilot is already built into their daily workflow. For businesses, by contrast, it is a clear mechanism for controlling spending: the AI assistant turns from a vague line item in a subscription into a managed resource with limits, forecasting and the ability to compare the effectiveness of different usage scenarios.

On a broader level, GitHub is merely cementing a trend already visible across the market for generative development tools. The more powerful the models and the more agentic features a product has, the harder it is to maintain the illusion of unlimited access at a fixed price. Vendors are gradually separating basic capabilities from resource-intensive modes, then tying the latter to credits, bundles or separate plans. That is less comfortable for users, but it shows the real cost of AI far more honestly.

The main takeaway is simple: GitHub is not just updating Copilot's price list but moving the product into a more mature commercial phase. The service's value will now be judged not only by the quality of its suggestions but by how well each credit spent is justified. If the new scheme proves transparent and predictable, the market will accept it as the norm. If not, teams will count return on investment more strictly and compare Copilot with alternatives whose limits are easier to understand.