GitHub Copilot moves to AI Credits on June 1. Here's what finance should do first.
/—On June 1, 2026, every GitHub Copilot plan moves to usage-based billing. Premium request units are gone. In their place: GitHub AI Credits, metered by token consumption — input, output, and cached — at each model's published rate, where one credit equals one cent. Seat prices don't change. What a seat buys you does.
For finance, this is the moment the AI coding line item stops behaving like SaaS and starts behaving like cloud.
What actually changed
Under the old model, every Copilot request cost the same regardless of which model ran or how much work it did. A one-line autocomplete and an hour-long agent session were priced identically. That's over.
Under AI Credits, you pay for the compute consumed. Each plan includes a monthly credit allotment; exceed it and overage is billed automatically at published rates — and unless an administrator sets a limit, there is no hard cap.
Three consequences matter for the budget.
Agentic usage is the cost driver — not autocomplete. Code completions and Next Edit Suggestions stay free and unlimited. What burns credits is everything agentic: chat, multi-file reasoning, code review, and autonomous agent sessions. The teams treating Copilot as an unlimited intern will find the ceiling first.
Model choice is now a budget lever. Premium models consume more tokens per task and carry higher rates. The same request costs different amounts depending on which model a developer reaches for — and most developers have no idea they're making a spending decision every time they switch.
Your bill is now variable. Last month's invoice no longer predicts next month's. Forecasting moved from arithmetic to modeling.
What to do before June 1
GitHub has published a preview billing experience. Use the window before the switch.
- Pull your most recent usage report. From the premium request analytics page, request the usage report (delivered by email as a CSV), then run it through the preview bill to see a model-by-model breakdown of where your credits would have gone under the new rates.
- Set a spending limit. Without one, overage bills automatically with no ceiling. A limit converts an open-ended liability into a known, capped number you can defend in a budget review.
- Identify your heaviest agentic workflows. The CSV will show which teams and which models drive consumption. That concentration is where your variance lives — and where governance pays for itself.
- Decide your annual-vs-monthly posture. Annual plans are being retired; if you're on one, you stay on premium requests until it expires, with model multipliers changing on June 1. Model the cost of riding it out versus converting to monthly usage-based billing now.
Why this lands on finance's desk, not just engineering's
A flat per-seat fee was a procurement decision you made once a year. A metered, model-dependent, agent-driven bill is a recurring forecasting problem — the same shape as cloud spend, and it will grow the same way if nobody owns it.
The organizations that handle this well won't be the ones that cap usage hardest. They'll be the ones that can see it: spend by team, by model, by workflow, with a forecast attached — so the conversation with engineering is about value per credit, not a surprise line item after the fact.
That visibility is the whole game now. June 1 just made it unavoidable.