GitHub pushes usage-based pricing

GitHub is changing how Copilot gets paid for. That sounds like billing plumbing, but it’s really a product story about what AI coding tools have become. Copilot is no longer just autocomplete and a quick chat box — it now runs longer, agent-style workflows across whole repos, and GitHub says the old pricing model stopped matching the actual compute bill. So on June 1, 2026, every Copilot plan moves from request counting to usage-based billing built around GitHub AI Credits. (github.blog) ### What actually changed? GitHub said all Copilot plans will stop using premium request units and start consuming AI Credits instead. Those credits are tied to token usage — input, output, and cached tokens — and the amount burned depends on which model you pick. GitHub also says base subscri(github.blog)e stays $39 per user. (github.blog) ### Why ditch request counting? Because a “request” stopped meaning anything consistent. A short chat prompt and a long autonomous coding run could be billed the same way even though the backend cost is wildly different. GitHub’s explanation is pretty direct — Copilot has shifted from an in-ed(github.blog)inference costs. (github.blog) ### What is an AI Credit? Basically, it’s GitHub’s house unit for metered AI usage. GitHub’s docs peg 1 AI credit at $0.01, and then publish per-million-token rates by model. So the bill now depends on two things: how many tokens your workflow uses and how expensive the chosen model is. Lightweight models cost less. Frontier models cost more. That’s the whole logic of the switch. (docs.github.com) ### What stays included? Not everything becomes metered. GitHub says code completions and Next Edit Suggestions remain included across plans and do not consume AI Credits. The metered bucket covers things like Copilot Chat, Copilot CLI, Copilot cloud agent, Copilot Spaces, Spark, and third-party coding ag(docs.github.com)ence while charging more precisely for heavier agent behavior. (github.blog) ### Why are customers nervous? Because usage pricing is fairer in theory but harder to budget in practice. GitHub is clearly aware of that. It’s rolling out a preview bill experience in early May so users can compare current spend with estimated spend under the new system, and it offers downlo(github.blog)uestion is not “is this elegant?” but “what is this going to cost me next month?” (github.blog) ### Is this only about Copilot chat? No — and that’s the bigger tell. GitHub also said Copilot code review will start consuming GitHub Actions minutes on June 1 for private repositories, on top of AI Credits. That means some agentic features now trigger two meters at once: model usage and work(github.blog) watch. (github.blog) ### Why now? Turns out GitHub had already started tightening the screws. On April 20, it paused new self-serve sign-ups for several individual Copilot plans, tightened usage limits, and adjusted model availability. GitHub framed those moves as a reliability fix (github.blog)e monetization half of that same reset. (github.blog) ### Bottom line This is GitHub admitting that AI coding tools are becoming compute businesses. Flat subscriptions worked when Copilot behaved like autocomplete. They work a lot worse when the product starts acting like an always-on software agent. Usage-based pricing makes the economics cleaner for GitHub — but it also shifts more planning, and more bill shock risk, onto customers. (github.blog)