What changed
GitHub used the last few days to make one thing unmistakable: GitHub Copilot is no longer being priced like a lightweight chat add-on. It is being priced like agent infrastructure.
On April 27, GitHub announced that all Copilot plans will move to usage-based billing on June 1, 2026. Instead of counting premium requests, Copilot will meter input, output, and cached tokens as GitHub AI Credits. Base subscription prices stay the same, but usage above the included allowance can now turn into extra spend. GitHub also said Copilot code review will consume GitHub Actions minutes on private repositories, on top of AI Credits.
Then, in an update published on April 29, GitHub explained why it had already started tightening the screws on individual plans: pausing new sign-ups, reducing some limits, and changing model access. The reason is straightforward. Agentic workflows are consuming far more compute than the old flat-fee structure was built for.
Why it matters
This is bigger than a pricing tweak. It is a product signal that the economics of AI coding tools are changing.
For developers, the practical shift is that agentic work now has an explicit marginal cost. A quick completion stays cheap, but longer cloud-agent sessions, larger context windows, and model-heavy review loops are becoming metered workflows. GitHub's own documentation makes that concrete: code completions and Next Edit suggestions stay included, while tools like Copilot Chat, Copilot CLI, Copilot cloud agent, Copilot Spaces, Spark, and third-party coding agents draw from the credit pool.
For teams, Copilot code review is the sharper change. If you enable automatic review on private repositories, those runs will start using GitHub Actions minutes beginning June 1. That means engineering managers now need to think about Copilot not just as a seat license, but as a workload that touches both AI and CI budgets.
What to do now
- Audit where you rely on Copilot code review in private repositories, because that feature is about to show up in Actions usage.
- Recheck model defaults and agent usage patterns, especially if your workflow leans on long-running or multi-file tasks.
- Treat June 1 as a budgeting change, not just a feature update. If your team adopted Copilot assuming flat-rate behavior, that assumption is now outdated.
- If you build developer-facing products, pay attention to the broader lesson: the market is starting to separate cheap autocomplete from expensive autonomous work.