You've probably already seen the number. Past nine hundred downvotes on the GitHub Community thread underneath GitHub's own announcement that Copilot is moving to "AI Credits." Underneath it, The Register ran "Angry devs vow to flee GitHub Copilot as metered billing takes hold." Somewhere in your feed this week was a screenshot of someone's credit meter draining to zero before lunch.
I'm not going to tell you that's overblown. Some of it is completely earned — if you're an org admin watching an agent chew through a seat's monthly allowance in an afternoon, that's a real budget problem and this piece isn't for you. I'm writing for the person who read that thread, felt the floor drop, and hasn't yet checked one specific thing: whether any of it touches what you actually do in Copilot all day, which for most people on Free or Pro is mostly autocomplete.
The one sentence that answers it
Here it is, taken straight from GitHub's own pricing page, word for word:
"Code completions and next edit suggestions don't use credits. They remain unlimited with every paid plan."
That's not a summary. That's not me softening it for you. That's the actual sentence, and it names your feature by name — completions, and Next Edit Suggestions, the little "here's the next change" prompts Copilot shows you as you type. Neither one touches the credit pool. On every paid plan, they're unlimited, exactly like before June 1.
If you're on Free, there's a wrinkle worth knowing precisely because it's not the same sentence: Free caps you at 2,000 completions and 50 chat requests a month. That cap isn't AI Credits either — it's a separate, older, flatter limit that predates this whole system. So even the exception has an exception, and the only way to know that is to have read past the headline sentence into the row that actually says "Free."
Notice what I didn't do to find that sentence. I didn't read the community thread's replies looking for consensus, past-nine-hundred-downvotes and all. I didn't read a recap site's summary of the summary. I went to GitHub's own plan page and searched the word "completions," because that's the word for the thing I actually do all day. It showed up twice, in two different official pages, worded identically both times — which is itself a small, useful check: a real policy sentence tends to survive being quoted in more than one place without drifting.
What the credit pool is actually for
The credits are real, and they're not nothing. GitHub's own docs put it plainly: usage-based billing started June 1, 2026, and the features that draw from it are the ones running a full model round-trip on your behalf — Copilot Chat, Copilot CLI, the Copilot cloud agent, Copilot Spaces, Spark, and third-party coding agents wired into Copilot. Every one of those sends tokens to a model and gets tokens back, and GitHub prices that at 1 AI Credit per cent, converted from the published per-model API rate for whatever you're running.
Completions and Next Edit don't do that. They're a much smaller, cheaper, purpose-built model doing one narrow job — predicting the next few tokens near your cursor — not a general chat model reasoning over your whole request. That's the actual mechanism behind the sentence: it's not a marketing carve-out, it's a different kind of request being billed a different way. Which is also why the sentence has held since June and nobody's found the fine print that quietly contradicts it — it's describing something structurally true about how the feature works, not a promise someone could walk back with an asterisk.
So if your day is mostly typing and accepting suggestions, with the occasional chat question, you are, on paper, close to unaffected by the thing that filled your feed with panic screenshots this week.
Why this matters more than this one changelog
Here's the part I actually want you to keep, because you will be back in this exact spot with a different tool before the year's out. Every credits-or-usage-limit announcement is written for two different readers at once: the individual on a $10 seat, and the org admin managing fifty of them and a real invoice. Most of the discourse you'll hit first — the downvoted thread, the outrage piece, the "here's what changed" recap — is written from the second reader's chair, because that's where the money and the drama both live. You are very often the first reader, and the FAQ you land on wasn't built with you in the room.
The fix isn't reading the whole changelog top to bottom and hoping the right line finds you. It's smaller than that: open the primary announcement — not the recap of it, not the outrage thread quoting it — and search for the exact name of the one feature you personally use. Not "credits," not "billing." The feature name. "Completions." "Chat." Whatever verb describes the thing you do forty times a day. Find the sentence that puts that word next to "does" or "doesn't," and stop reading once you have it. That sentence is the whole job. Everything else in the announcement is written for someone else's seat count.
Sure, but — doesn't that mean you're just trusting the vendor's own framing? Not exactly. You're trusting a specific, falsifiable sentence that names a feature, not a vibe about the company's intentions. If GitHub's completions sentence turns out to be wrong, that's checkable — people would notice their completions suddenly metered, and they'd say so loudly, the same way they said so loudly about credits. A specific claim you can watch fail is a different thing from a policy summary you're taking on faith.
The proof this isn't a one-time check
You're reading this on September 2. The other half of this story landed yesterday, and if you weren't looking for it, you'd have missed it: the elevated AI Credit allowance GitHub gave existing Copilot Business and Enterprise customers as a launch cushion — 3,000 credits a seat on Business, 7,000 on Enterprise, running June 1 through September 1 — reverted on schedule. As of yesterday, Business is back to the standard 1,900 credits per seat, Enterprise to 3,900, per GitHub's own published rates. That's a real cut for the org side of this, and if you manage seats, it already hit your account while you were reading a different thread.
If you're the reader this piece is for, though, notice what that news isn't. It's not a Free or Pro plan change. It doesn't touch your seat, your allowance, or the completions sentence that still holds from June. The habit is exactly the one you just used: before you let a fresh wave of Copilot headlines change how you feel about the tool, find the line that says which plan it's actually talking about. This one names Business and Enterprise explicitly. If that's not your seat, it's not your news — and knowing that took one sentence, not a re-read of the whole thread you already doomscrolled once this week.
Why this matters
The GitHub thread earned its downvotes. The Register headline isn't wrong. What's wrong is treating either one as the whole answer for you specifically, when the actual answer was one sentence away in GitHub's own announcement the entire time — and every vendor that ships usage-based pricing after this one is going to bury an equally exact sentence in an equally long page. The failure mode this habit prevents isn't "you got the facts wrong." It's "you made a real decision — switching tools, changing how you work — off a headline written for someone else's bill."
One thing to actually do
Next time a tool you use announces credits, tokens, or usage limits — and there will be a next time — don't start with the outrage thread and don't start with someone's recap. Open the vendor's own announcement, Ctrl+F the exact name of the one feature you personally touch every day, and read only the sentence that pairs your feature with "does" or "doesn't." That's the whole check. It takes less time than the panic did.