Copilot AI credits: how GitHub now bills by the token

Copilot AI credits are GitHub's billing unit for Copilot since June 1, 2026: each credit is worth one US cent, and you spend them on the tokens a model reads and writes, priced at that model's API rate. The change ended the flat per-message counting of premium requests, which means a long agent run now costs more than a one-line question.

Why GitHub moved off requests

GitHub announced the switch on April 27, 2026. Its reasoning was blunt: under premium requests, a quick chat question and a multi-hour autonomous coding session could cost a user the same. As agents took on longer tasks, that pricing stopped matching what the models cost to run. Tokens fix the mismatch, because the bill now follows the work.

Premium requests had been billed since June 18, 2025, with a multiplier per model. That system lasted under a year for monthly plans.

What spends credits and what does not

Credits cover input, output and cached tokens. Picking a larger model costs more per token, so model choice now shows up directly in the balance.

The rules that matter day to day

• Reset: the monthly allowance refills at 00:00 UTC on the first day of each calendar month.
• No rollover: unused credits are lost at the reset.
• No fallback: GitHub removed the old behaviour of dropping you to a cheaper model when you ran out.
• When it is gone: you can upgrade, pay for more usage, or wait for the reset. Completions keep working on paid plans.
• Budgets: individuals can set a dollar budget for extra usage; organizations can set budgets at enterprise, cost center and user level, and choose to allow more usage or cap spend.

Annual Pro and Pro+ subscribers are the exception. They stay on premium requests and multipliers until their plan expires, then move to Copilot Free unless they switch to a paid monthly plan.

GitHub's docs give each plan its own monthly allowance, and those figures have already moved since the announcement. Check your own plan page rather than a number quoted in an article.

Seeing where the credits go

GitHub's billing page, under Metered usage, is the record of what you spent, and a downloadable CSV lists the model behind each request. Editors also show a summary behind the Copilot icon.

Developers who run Copilot alongside Claude Code, Codex or Cursor have a different problem: four tools, four dashboards. UsageAtlas, a free and open-source desktop tracker for Windows, macOS and Linux, puts them in one view. For Copilot it shows the quotas and seat credits GitHub reports, with the reset time where one is supplied, and leaves a limit marked unknown when GitHub does not send it. It does not read prompts, and local tracking needs no account. The UsageAtlas guide to Copilot credits covers the setup. For a per-model cost breakdown, GitHub's CSV is still the better source.

What to watch

The move puts Copilot on the same token-based footing as most AI coding tools. The practical effect is that heavy agent users will notice their balance in a way they did not under request counting, and model choice becomes a cost decision.

FAQ

What is one Copilot AI credit worth? GitHub's docs set one AI credit at $0.01 USD.

Do Copilot code completions use AI credits? No. Completions and next edit suggestions are not billed in credits on paid plans.

Do Copilot AI credits roll over? No. The allowance resets on the first of each month and unused credits do not carry over.