What changed
On April 27, GitHub announced that all GitHub Copilot plans will move from premium request units to token-based "GitHub AI Credits" on June 1, 2026. Instead of counting requests, Copilot usage will now be metered by input, output, and cached tokens, using per-model pricing published in GitHub Docs. GitHub says the change reflects how Copilot has shifted from short chat interactions to longer, agentic workflows that consume meaningfully more compute.
Source: GitHub announcement, models and pricing.
Why it matters
This is a product shift, not just a billing tweak. For developers, Copilot is becoming a metered AI tool where workflow design and model choice now affect cost directly. GitHub’s docs say code completions and Next Edit suggestions remain included for paid plans, but features like Copilot Chat, Copilot CLI, cloud agent sessions, Spaces, Spark, and third-party coding agents will consume AI Credits.
That makes Copilot feel closer to an API-backed AI platform than a flat SaaS subscription. If you rely on long-running agentic sessions or frontier models, budgeting and usage visibility become part of normal developer operations.
Source: usage-based billing docs, prep guide.
The second bill teams may miss
GitHub also published a separate April 27 changelog entry for Copilot code review. Starting June 1, each code review will be billed in two ways on private repositories:
- AI Credits for the model usage
- GitHub Actions minutes for the runner time behind the review workflow
GitHub says public repositories are unaffected because Actions minutes remain free there, and self-hosted runners avoid GitHub-hosted runner minute consumption. For teams that adopted Copilot code review assuming it lived entirely inside the Copilot subscription, this is the practical surprise to plan for.
Source: Copilot code review billing change.
What developers should do now
- Check which Copilot features in your workflow actually consume credits.
- Review model pricing, because lightweight and frontier models now have visibly different cost profiles.
- If you use Copilot code review on private repos, forecast the extra GitHub Actions minutes before June 1.
- Use GitHub’s billing preview tools in early May to compare April usage against the new credit model.
For portfolio-site readers building with AI every day, the important signal is simple: AI coding tools are maturing from flat-fee helpers into usage-metered infrastructure. That changes how you choose models, how aggressively you automate, and how you explain tool spend to clients or employers.