GitHub quietly changed the economics of the most widely used AI coding tool on June 1, and the change rewards paying attention. Copilot no longer counts "premium requests." It bills in AI Credits, where one credit is one cent of model usage, and it added a new $100/month Max tier for people running agents all day. Two facts decide whether this is good or bad news for you, and they point in opposite directions.
The good news first, because it's the part most coverage buries: inline code completions and next-edit suggestions are now free on every paid plan and consume zero credits. The feature the majority of developers use the most — grey-text autocomplete — is no longer metered at all. Credits are spent only by chat, agent mode, code review, and the Copilot CLI. If you're a completion-first coder, your effective bill just went down.
The trap is on the other side of that line. Agent mode — where Copilot plans, edits across files, and runs autonomously — burns credits far faster than a chat message, and a long autonomous run can eat a lower tier's allowance in an afternoon. That's the whole reason Max exists.
The tiers, priced in what you actually get#
Because a credit equals a cent of model usage, the plans are easy to compare on value received:
- Pro — $10/month, 1,500 credits (~$15 of usage). For developers who mostly accept completions and dip into chat occasionally. Completions are free, so most of this pool goes untouched.
- Pro+ — $39/month, 7,000 credits (~$70 of usage). The balanced pick for regular chat plus occasional agent runs.
- Max — $100/month, 20,000 credits (~$200 of usage). Built for sustained, agent-driven work. The 20,000 is 10,000 base plus 10,000 "flex."
Note the pattern: every tier hands back more model usage than its sticker price — Pro is ~1.5×, Max is ~2×. Past the included pool, additional credits bill to your card on file at a penny each, so overages are predictable but uncapped by default.
The decision isn't really about price. It's about which side of the free line your work sits on. Completions are free; autonomy is metered. Size your plan to how much you let the agent drive, not to how many hours you spend in the editor.
How a solo founder should choose#
Don't guess your burn — measure it. The usage-based model means the right answer is legible in your own credit meter.
If you code by accepting completions and reach for chat now and then, stay on Pro. You're paying $10 mostly for the chat you occasionally use; the thing you use constantly is free.
If chat and agent mode are part of your daily loop but you're not running long autonomous jobs, Pro+ at 7,000 credits is the middle that keeps you clear of overages without over-buying.
If you run agents for hours a day — parallel PRs, big refactors, overnight tasks — and you were watching a Pro+ allowance evaporate into overage charges, Max is simply cheaper than paying penny-per-credit past 7,000. At that volume, the $100 flat is the frugal choice, not the extravagant one. And if the meter still stings, remember the credit pool is model-priced: routing agent work to a cheaper backend changes the math, which is the whole premise behind swapping Copilot's model for an open-weight one and behind picking a coding-agent backend on cost, not brand.
None of this settles the tool question — whether Copilot, Cursor, or Claude Code deserves your primary seat is a separate comparison. But it does settle the plan question: match the tier to your metered burn, and stop paying for autonomy you don't use.
The honest move is to run two weeks on your best guess, open the usage dashboard, and look at where your credits actually go. Under request-based pricing you couldn't see that. Under credit-based pricing you can — which, whatever you think of the new bill, is the part that lets you stop overpaying.
