GitHub Copilot Switches to Usage-Based AI Credits Billing

As of June 1, 2026, GitHub Copilot runs on usage-based billing. The old system counted “premium requests” against a monthly cap. The new one meters what you actually use and charges it against a pool of GitHub AI Credits, and monthly plans all migrated over on June 1. For annual plans, GitHub’s billing docs state that existing annual Pro and Pro+ subscribers keep premium-request billing until the plan ends, then are automatically downgraded to Copilot Free, with the option to upgrade to a monthly paid plan early and receive prorated credits, or cancel for a prorated refund. If you only ever used Copilot for autocomplete, nothing about your bill changes. If you lean on chat and agents with frontier models, the math is now very different.

Here is what moved, and what to do about it.

From Premium Requests to AI Credits

The previous model gave each plan an allowance of premium requests, with some models counting as “free” and others burning the allowance faster. That whole premium-request allowance is gone now, and in its place there’s one unit: the AI Credit, where 1 credit equals $0.01.

Credits are metered by token consumption, input, output, and cached tokens, at each model’s underlying API rate. A short prompt against a small model costs a fraction of a credit. A long agent run against an Opus-class or other frontier model costs many. The bill now tracks the real cost of the work rather than a flat request count.

One important carve-out: code completions and next-edit suggestions stay unlimited and free on every paid plan. Credits are drawn down by chat, agents, the Copilot CLI, and code review (code review also consumes GitHub Actions minutes). For a lot of developers, that completions line is the whole story.

What the Plans Cost Now

The seat prices themselves did not change at all here. What’s new is the credit allowance that comes attached to each tier.

• Free: limited completions and chat, plus a small monthly credit allotment.
• Pro ($10/month): unlimited completions, plus a monthly credit allotment GitHub’s docs put at 1,500 credits.
• Pro+ ($39/month): unlimited completions, plus a larger monthly allotment.
• Max ($100/month): unlimited completions, plus a 20,000-credit monthly allotment (10,000 base + 10,000 flex) for heavy individual use.
• Business ($19/user/month): team features, with credits pooled across the organization.
• Enterprise ($39/user/month): org codebase indexing and custom models, with pooled credits.

When you run past the allotment, paid plans can set a dollar budget for additional usage so work continues instead of cutting off mid-task. Organizations get pooled credits, so one heavy user can draw from capacity a lighter teammate did not spend.

Who Actually Feels This

Light and moderate users are mostly insulated. Completions are free, and a $10 Pro allowance covers a fair amount of everyday chat and agent work. The people who will notice are heavy agent users running long, multi-step tasks against the most expensive models all day. That is exactly where token-metered billing diverges most from a flat request count.

If you want to see how this sits against the rest of the market, our AI coding tools pricing comparison breaks down Copilot’s credits alongside Cursor’s request pools, Claude’s shared usage, and raw API pricing, including why API billing tends to run far more expensive than a subscription for the same work.

What to Do Now

Set a usage budget if you run agents seriously, so a runaway task does not quietly rack up overage. Keep an eye on which model your agent defaults to, since frontier models are where credits go fastest. And if you mostly use Copilot for inline completions, you can stop reading here, that part is still free.

Related Guides

• AI Coding Tools Pricing Comparison, full breakdown of every tool’s pricing model, including Copilot’s AI Credits
• Best AI Coding Tools, overview of every major tool and where each one fits