On June 1, 2026, GitHub flips every Copilot plan — Business, Enterprise, even the legacy Individual seats — onto a new model: a fixed monthly seat charge plus a metered allotment of "GitHub AI Credits" that get consumed by every Copilot call.
For most teams this lands somewhere between "shrug" and "what just happened to our bill." The shrug case is teams that use Copilot lightly. The shock case is teams that lean on Copilot Chat, Workspace, and the longer-context models — that usage now meters, and the meter doesn't stop at the old flat-seat price.
Here's what's changing in plain English, who feels it most, and three things to do this week so the June invoice doesn't surprise you.
The model — before and after.
The old Copilot Business plan was the cleanest line item on most engineering AI budgets: $19/seat/month, unlimited completions, unlimited chat. The bill scaled with headcount, not with use. A heavy user and a light user cost the same.
Starting June 1, every seat still has a fixed monthly charge, but it now includes a monthly allotment of AI Credits instead of unlimited use. Cross the allotment and you start paying per request, with different rates for different models (Claude Sonnet, GPT-4.x, the longer-context variants). Cached tokens are cheaper than fresh ones.
Flat $19/seat.
Unlimited Copilot completions and chat. The seat count is the only variable. A 10-person team paid $190/mo regardless of whether they used Copilot once a week or 500 times a day.
Fixed seat + AI Credits.
Each seat still has a fixed price, but it includes a credit allotment. Calls eat credits at different rates per model. Past the allotment, every additional request bills. Two seats with very different usage now cost very different amounts.
This isn't unique to GitHub. The broader trend in dev-AI subscriptions is toward metered pricing, because the underlying inference cost varies enormously per user — a flat seat price means heavy users get subsidized by light ones, and vendors are walking away from that subsidy.
Who gets hit hardest.
The teams most exposed share two characteristics: heavy chat use and long-context models. Lots of inline completions on a few short files is cheap; constant chat sessions against multi-thousand-line repos is expensive.
The table below is a thought experiment, not data. We don't have insight into anyone's actual Copilot bill — we're sketching the shape of the change so you can build your own estimate from your own usage. Run the math against your team's real numbers before assuming you'll fit any of these buckets.
The exact dollar depends on which models the team picks, how much usage is cached, and whether the org is on Business or Enterprise (Enterprise gets a larger included credit allotment per seat). The point isn't to predict a specific number — it's to recognize the shape of the change. Flat-seat pricing was a subsidy for power users. That subsidy is going away. Light users break roughly even; everyone else pays more in proportion to how much they use Copilot.
Three things to do this week.
None of this requires buying anything. All of it requires looking at numbers you may not have looked at before.
1. Pull your last 30 days of Copilot usage from GitHub.
The Copilot Admin dashboard now shows per-seat usage broken out by editor and feature. As an org owner, log into github.com → your organization → Copilot → Usage. Export it. Specifically look at:
- Active vs. inactive seats. A seat that hasn't fired a call in 14 days is one you can probably reclaim — yank it now and skip a billing cycle. Inactive-seat rates vary wildly across teams; the only way to know yours is to look.
- Chat-heavy users. The ones who'd cost the most under the new model. Identify them, not to punish them, but to predict the bill.
- Model selection. If your team is defaulting to Claude Opus or GPT-4 Turbo for everything, switching the routine work to a cheaper default model can materially cut credit burn — the lighter models are priced well below the premium ones.
2. Decide which power users get the Enterprise tier.
Enterprise's per-seat AI Credit allotment is larger than Business. For one or two genuine power users on a small team, paying the Enterprise differential for those seats and keeping the rest on Business often comes out cheaper than putting everyone on Business and watching the power users blow through their credits. Run the math against your actual usage from step 1.
3. Set up a forecast — once — before June 1.
Take your last 30 days of usage data, multiply by GitHub's new published per-call rates, add the fixed seat cost, and you have a forecast for June. It'll be wrong (usage shifts in summer, teams onboard new tools, the credit rates may tune in the first few weeks) but a wrong forecast you've thought about beats a surprise on July 1.
Build the forecast somewhere durable — a spreadsheet that you can update monthly, or a tracking tool that ingests both the API bills (OpenAI, Anthropic) and the subscription tools (Copilot, Cursor, ChatGPT Business) in one place. The shift Copilot is going through is the same shift OpenAI, Anthropic, and most other dev-AI vendors have already made: metered, model-dependent pricing that's harder to predict from headcount alone.
The bigger pattern.
Three things are happening to the AI-tools-for-engineers market at once: metered pricing is replacing flat-seat pricing, the gap between light and heavy users is widening, and the bills are landing in different inboxes (finance pays Copilot, engineering pays the API). The teams that handle this well will be the ones who can see all of it in one place, not the ones who have the cheapest plan.
If you want help with that — the "all of it in one place" part — Stax is building exactly that: a single dashboard that pulls Copilot usage, API spend, and the other AI subscriptions into one monthly view. Flat $49/mo, no usage-based escalation. Pre-launch — join the waitlist to get early access.
Built and read this far? Stax is the dashboard that does step 3 automatically for Copilot, OpenAI, Anthropic, Cursor, and the rest. Flat $49/mo, no metering, no surprise rows.
Join the waitlist →