GitHub Copilot’s Usage Billing Is Hitting Enterprise Teams

GitHub Copilot’s usage-based billing went live on June 1, and enterprise engineering teams are now dealing with a very different cost model. Instead of paying a fixed per-seat fee, customers are now using credits that are consumed as AI features run, according to the article and GitHub’s billing changes.

The shift matters most for teams that use Copilot for more than simple autocomplete. Code completions and Next Edit Suggestions remain free and unmetered, but chat, agent mode, multi-step sessions and tool calls are now billed through GitHub AI Credits, which are calculated from token consumption. Copilot code review also consumes GitHub Actions minutes in addition to credits.

That has left some engineering leaders trying to figure out how much their AI bill will vary from week to week. The article says several leaders at the Gartner Summit were asking the same question: what happened to the Copilot bill?

The complaints are not just about higher prices. They are about unpredictability, especially for agentic workflows where developers ask the tool to take multiple steps, call tools and keep running until a task is done. One Pro+ user reported burning through roughly 8% of a monthly allotment in two hours, while another said a single change request cost more than $6.

Another example cited in the article involved Claude 4.8 being used to fix site issues, which consumed 1,180 credits, or about 16% of a Pro+ monthly allowance. A separate file review, with no code changes, reportedly used 20% of one developer’s monthly allowance. The article also says people are circulating projections of monthly costs rising sharply, including from $29 to $750 and from $50 to $3,000 in heavy agentic workflows.

GitHub says the pricing is based on standard per-model API rates, and one commenter quoted in the article said the models cost “exactly the same price as direct from OpenAI and Anthropic.” The article argues that the subsidy was the flat subscription, and that the new billing model simply exposes what agentic coding actually costs.

That change has also made budgeting harder. GitHub’s Billing Preview tool was intended to estimate costs before the switch, but the article says it uses discounted credits, so it underestimates what enterprises will actually pay. GitHub also warns that older IDE and extension versions can show inaccurate pricing.

The broader concern is that Copilot is not the only product pushing enterprise AI usage toward consumption pricing. The article points to Uber as an example of how quickly AI tooling costs can grow at scale. According to The Information, Uber spent its entire 2026 AI coding tools budget in four months and later capped employee spending at $1,500 a month.

For engineering leaders, the article recommends pulling actual usage reports, modeling spend against real rates, setting spending caps and user-level budgets, and matching the model to the task instead of defaulting to the most expensive option. It also warns against relying on a single vendor’s pricing or model availability for the whole workflow.

GitHub Copilot’s new billing model is already live, and the people dealing with it now are enterprise teams.