For its first few years, GitHub Copilot worked like Netflix: pay a flat monthly fee, use as much as you like. On 1 June 2026 that ended. Every Copilot plan now runs on usage-based billing through a new unit called GitHub AI Credits, consumed in proportion to the tokens your interactions use.
GitHub announced the change on 27 April 2026 and was unusually direct about why. Copilot, it said, is not the same product it was a year ago. It now powers complex, agentic workflows that consume far more compute, and pricing had to align with actual cost.
Key takeaways:
- Announced 27 April 2026; live for all plans on 1 June 2026.
- Premium request units (PRUs) are retired. Usage is now measured in GitHub AI Credits based on input, output and cached tokens at each model's listed API rate.
- Every plan includes a monthly allowance of credits; paid plans can buy more or set a hard budget cap.
- Code completions and Next Edit Suggestions remain unlimited. Chat, code review and agent tasks consume credits.
- Copilot code review now also consumes GitHub Actions minutes, because it runs as an agent on Actions.
- Annual plans were retired; monthly plans migrated automatically.
How the old system worked, and why it broke
Under request-based billing, each interaction with a premium model cost one "premium request unit", multiplied by a factor depending on the model. A quick chat question and a forty-minute autonomous agent session that touched two hundred files cost the same number of requests. That was fine when Copilot mostly autocompleted lines. It stopped being fine when developers began delegating whole tasks to agents that loop, read, plan and rewrite for extended periods.
The mismatch is the whole story: a flat fee per request cannot survive when the compute per request varies by three orders of magnitude.
How the new system works
| Element | Detail |
|---|---|
| Unit | GitHub AI Credits, denominated in US dollars |
| What consumes credits | Chat, code review, agent tasks and other premium features |
| What does not | Inline code completions, Next Edit Suggestions |
| How usage is measured | Input tokens + output tokens + cached tokens, at the published API rate for the model used |
| Overage | Paid plans can purchase more credits or set a spending limit that stops usage when reached |
| Example plan | Copilot Pro+ at $39/month includes $39 of monthly credits |
Third-party analyses report one credit as equal to $0.01 and give per-plan credit allowances; treat those figures as secondary and check GitHub's billing documentation for current numbers.
What changed for teams
Engineering managers now have a variable line item where they used to have a fixed one. GitHub added controls to manage it: a preview bill in May before the switch, per-user budgets that act as hard stops, and an ai_credits_used field in the Copilot usage metrics API (added 19 June 2026) so organisations can see consumption per developer per day.
Later in the summer GitHub tightened further. Secondary reporting indicates new sign-ups for Pro, Pro+ and Max plans were paused from 2 August, and that Business and Enterprise seats move to prepaid billing from 1 October 2026. Confirm both on GitHub's changelog before relying on them.
Concept to take away: tokens are the unit of cost. Every large-language-model product ultimately bills by tokens, whether or not the price list says so. Learning to estimate token usage (roughly three-quarters of a word per token in English; far more for code and long contexts) is now a practical engineering skill, like estimating cloud storage costs.
Why learners should care
This is the clearest signal yet that the economics of AI-assisted development have shifted. "Unlimited AI" was a launch-phase subsidy. As agents do more of the work, expect metered pricing to become the norm across tools, which makes efficiency (smaller prompts, better context management, choosing the right model tier) a real professional advantage.
