On June 1, 2026, GitHub Copilot moved entirely to usage-based billing. The old Premium Request Units (PRUs) were replaced by a new unit: GitHub AI Credits, priced on the tokens each interaction consumes (input, output, and cached tokens) at each model's published API rates. Code completions and Next Edit Suggestions remain unlimited, but everything else — agentic workflows in particular — is now metered. Each plan includes a monthly allotment of AI Credits, with paid plans able to buy more; Copilot Pro+ is $39/month and includes $39 of AI Credits.
1. Why It Exploded Into Debate
Two things made this change controversial. First, the "downgrade fallback" safety net is gone: previously, when you exhausted your premium allotment, Copilot would quietly fall back to a cheaper model so you could keep working; that buffer is gone, and going over your allotment now bills directly. Second, heavy agentic users may see bills jump 10x to 50x — because an agent often reads many files and reasons over multiple turns per task, consuming far more tokens than plain completions. Multiple communities and outlets reported "meter shock" on June 1 itself.
2. The Hidden Actions Cost
On top of that, from June 1, reviewing a PR with Copilot counts against your Actions minutes at the same per-minute rates as any other Actions workflow — a cost that is easy to overlook when budgeting.
My Take
My read: this isn't GitHub getting more expensive or worse — it's AI-assisted development finally being aligned to real cost. In the flat-fee era, token costs were absorbed and subsidized by the platform, so usage felt free; usage-based billing puts the economics of every agent task on the table. That's healthy, even if the transition stings. For teams, three concrete moves. First, measure before you panic: use the built-in usage tracking to find which people, repos, and task types are burning credits — odds are 20% of agent tasks eat 80% of the cost. Second, make model selection a discipline: cheap models for completions and small fixes, premium models only for genuinely complex refactors and cross-file tasks, and write that rule into your team norms. Third, re-evaluate your tool mix: locally runnable open-source options (such as self-hosted models via Ollama) are now worth a serious cost calculation for privacy-sensitive or high-frequency, low-complexity tasks. Ultimately, usage billing forces us to run "AI writing code" as a production line with real costs rather than an all-you-can-eat toy — and the teams that build that cost awareness earliest will have the edge this year.
Sources
- GitHub Copilot is moving to usage-based billing — The GitHub Blog
- GitHub Copilot is moving to usage-based billing — community Discussion #192948
- GitHub Copilot Pricing Change Drives Backlash: Agentic Bills Jump 10x to 50x — TechTimes
- GitHub Copilot's New AI Credits Billing: What Changes June 1 — FindSkill.ai
- GitHub Copilot · Plans & pricing