GitHub puts Copilot on the meter, and long agent runs now change the bill

GitHub said on April 27 that Copilot will shift all plans to usage-based billing on June 1, 2026. That is not a small packaging tweak. It is GitHub admitting that multi-step coding agents have outgrown the economics of flat request buckets, and that the era of “just leave the agent running” now comes with a visible meter.

The core change is straightforward. Premium request units disappear and GitHub AI Credits take their place. Credits will be charged on token consumption across input, output, and cached tokens, using per-model API rates. Base subscription prices stay where they are: Copilot Pro at $10 per month, Pro+ at $39, Business at $19 per user, and Enterprise at $39. But the included monthly usage now matches those dollar amounts instead of hiding behind abstract request counts.

What changes the day-to-day math is everything around that meter. GitHub says code completions and Next Edit suggestions stay included, but fallback experiences disappear. Under the old system, a user who exhausted premium requests could keep going on a cheaper model. Under the new one, the real question becomes budget: either there are credits left, or an admin has allowed extra spend. In a separate GitHub changelog post, the company also said Copilot code review will start consuming GitHub Actions minutes in addition to AI Credits, turning automated review into an explicit infrastructure cost line.

GitHub is offering cushions because it knows this will sting. Business customers get $30 in monthly included credits for June through August, while Enterprise customers get $70. Unused included usage can be pooled across an organization instead of stranded seat by seat, and admins will get budget controls at the enterprise, cost-center, and user levels. That is a real improvement for procurement teams that hated the old combination of opaque limits and unpredictable heavy-user behavior.

The deeper signal is strategic. GitHub says Copilot is no longer an editor sidebar with occasional chat, but an agentic platform that can run long, repository-wide sessions. Once that framing is accepted, usage-based pricing follows almost automatically. The next thing to watch is whether engineering teams throttle agent use once the bill is legible, or whether pull-request throughput gains are strong enough that finance teams decide the meter is worth it.