What Copilot’s New Billing Means for Your Bottom Line

As of June 1st, a developer who lives in agent mode can now spend more in one afternoon than their Copilot seat costs for the whole month.

Before June 1st the GitHub Copilot was easy math: pay the seat price, get your bucket of requests, and a quick “explain this function” cost the same as building out an entire feature. That part is over. GitHub moved Copilot to usage-based billing, and now you pay for the tokens your team actually burns. It’s now a variable cost, and it needs to be managed like one.

What Changed

GitHub dropped premium request units in favor of GitHub AI Credits.

• Seat Prices Held. Pro is $10 a month, Pro+ is $39, Business is $19 per user, Enterprise is $39. Each one includes a monthly credit allowance equal to its price.
• Autocomplete Is Still Free. Completions and Next Edit suggestions don’t touch your credits. The meter runs on chat, agent mode, code review, the CLI, and the cloud agent.
• The Fallback Is Gone. You used to drop to a cheaper model when you ran out and keep working. Now the metered features just stop until your next cycle.
• Watch The Fine Print. Credits don’t roll over, and code review now eats GitHub Actions minutes on top of credits.
• Pooled Credits. All of your AI Credits are now in a pool that every developer can pull from. That means power users can draw more when they need it.

Subscription, Meet Utility

Here’s the shift in one line: Copilot used to be a subscription, a flat fee for all you can eat. Now it’s a utility, metered and closer to your cloud bill than a software seat.

That brings the same headache your cloud bill already has. Usage you can’t see is usage you can’t control. Picture a developer living in agent mode. One serious agentic run on a premium model can burn through millions of tokens, and a handful of those a day will clear the $19 of credits in a Business seat well before the month is out. That doesn’t make the work bad. It just means the cost shows up whether or not anyone was watching for it.

The heaviest users are already reporting bills several times – sometimes ten times – what they paid under the flat rate. And here’s the part worth sitting with: the developers getting the most out of Copilot are the same ones running up the biggest tab. The answer isn’t to rein them in. It’s to see what they’re spending and plan around it.

For Leaders: Get Ahead of the Bill

GitHub did ship some very useful tools you need to enable:

1. Start With A Baseline. The billing dashboard under Billing & licensing → AI usage breaks consumption down by user, model, org, and cost center. Look at real numbers before you set any limits, or you’re just guessing.
2. Enable User-Level Budgets. Cap how much an individual user can consume per billing cycle, from both the shared pool and additional usage. A $0 USD user-level budget blocks the user immediately.
3. Enterprise Spending Limits. Cap total metered charges across your entire enterprise after the pool is exhausted.

A little runway while you do this: through August, existing Business and Enterprise customers get extra promotional credits, $30 and $70 a month on top of the existing per seat AI Credits. Use that window to find your real baseline before it disappears.

For Developers: The Model You Pick Is a Cost

• Match The Model To The Job. The lighter included models are free and perfectly good for everyday questions. Save the frontier models for the gnarly, multi-file problems where they earn it.
• Give Your Agents A Tight Scope. A specific task costs a fraction of an open-ended “go fix everything” that wanders the codebase racking up tokens. (Our RPI approach — research, plan, implement — exists for exactly this reason.)
• Glance At Your Dashboard Now And Then. A great month turns into a bad surprise when nobody’s looking at the meter.

Final Thoughts

This isn’t a Copilot quirk – Cursor and Windsurf already made the same move, and usage-based billing is now the default across every serious agent tool. Copilot didn’t get worse, and for most people it isn’t even more expensive. What changed is the shape of the cost: a flat subscription became a number that moves with how you use it. Anyone who’s managed a cloud bill knows how this story ends. The teams that set a baseline, capped their spend, and tracked it by team came out fine. The ones who assumed the number would hold still got their lesson at the end of the month.

Put the guardrails up early, and you keep the productivity without the sticker shock.