GitHub shifts to usage‑based AI billing

GitHub is changing what Copilot costs mean inside a company. Until now, most teams could think about Copilot as a seat license with some usage rules around the edges. Starting June 1, 2026, that changes — Copilot moves to usage-based billing, and GitHub is now giving admins an April report that previews what that new meter would have looked like in practice. ### What changed this week? On May 12, GitHub said Copilot Business and Copilot Enterprise admins can download a usage report showing how April activity translates into GitHub AI Credits. That matters because AI Credits become the new billing unit on June 1. The report is basically a dress rehearsal — one month of historical usage, re-expressed in the pricing system companies are about to live under. (github.blog) ### What are AI Credits, exactly? GitHub is replacing “premium requests” with a token-based unit. In plain English, Copilot will now meter the amount of model work being done — input tokens, output tokens, and cached tokens — and convert that into AI Credits using model-specific rates. So the bill depends less on how many times someone clicked a feature and more on how much model compute their workflows actually consumed. (github.blog) ### Why is GitHub doing this? Because one request is no longer one thing. A quick autocomplete, a long chat, a multi-step agent run, and a code review can all hit very different models and consume very different amounts of compute. The old request-based framing gets fuzzy once Copilot starts acting more like an agent than a simple assistant. Usage-based billing lets GitHub charge in a way that tracks the real underlying cost more closely. (docs.github.com) GitHub says that is part of making Copilot sustainable and reliable as heavier workflows become normal. ### Who needs to care first? Admins, finance teams, and engineering managers. The April preview report gives them a chance to spot which teams, features, or models are likely to drive spend before the switch flips. GitHub’s prep tools also let some users compare current billing with projected AI Credit costs side by side, including additional usage beyond included amounts. That means this is not just a product change — it is a budgeting change. (github.blog) ### What gets more expensive to think about? The agentic stuff. Longer sessions, more capable models, and workflows that chain multiple steps together all put pressure on included usage. GitHub has already been adjusting plan details for individuals in response to that reality, and it has flagged that code review will also consume GitHub Actions minutes starting June 1 on top of AI Credits. So some Copilot features are turning into stacked infrastructure costs, not just a flat add-on. (github.blog) ### Does this only hit enterprises? No. GitHub says all Copilot plans are transitioning to usage-based billing on June 1, though the mechanics differ by plan. Individual plans get monthly AI Credit allowances too, with paid tiers offering more room before extra charges kick in. But the enterprise story lands harder because companies have more users, more governance, and a bigger need to forecast spend before procurement gets surprised. (github.blog) ### What does this change inside companies? It makes AI legible to the budget. A seat license is easy to approve and then ignore. A variable usage bill gets watched. Once Copilot spend shows up as a metered line item, teams will start asking sharper questions — which models are worth it, which workflows save enough time to justify the cost, and whether some usage needs caps or policy controls. GitHub already documents budget and allowance controls for premium usage, and that logic becomes more important under AI Credits. (github.blog) ### So what’s the real takeaway? This is GitHub admitting that AI coding tools are no longer simple SaaS seats. They are becoming compute products with software wrappers. The April report is useful because it gives companies one month to see that future in spreadsheet form before June 1 makes it real. (github.blog) (docs.github.com)