GitHub Copilot Now Bills by Token: What the Switch to AI Credits Means for Your Workflow

As of June 1, 2026, GitHub Copilot no longer counts premium request units. Every plan has moved to usage-based billing, measured in GitHub AI Credits, priced by the token. The flat-rate model that made Copilot feel like an all-you-can-eat subscription is gone, at least for chat and agentic work.

Here is what actually changed and what it means in practice.

What Are GitHub AI Credits?

The old system tracked “premium requests” as a rough proxy for usage. One chat message was one request, more or less, regardless of how much compute it consumed. That worked when Copilot was primarily an in-editor chat tool, but it broke down as agentic sessions became common. A quick question and a multi-hour autonomous coding session are very different things in terms of inference cost, and the old model treated them identically.

AI Credits are the replacement. One credit equals $0.01 USD. When you send a message or kick off an agentic task, credits are consumed based on token usage: input tokens (what you send to the model), output tokens (what the model returns), and cached tokens (context the model stores or reuses). Each model has its own published rate, so using a more capable and expensive model costs more credits per interaction.

Credit allocations are matched to subscription price:

• Copilot Pro ($10/month): $10 in AI Credits
• Copilot Pro+ ($39/month): $39 in AI Credits
• Copilot Business ($19/user/month): $19 per user
• Copilot Enterprise ($39/user/month): $39 per user

Plans include two layers: base credits (the fixed monthly amount above) and a flex allotment, an additional buffer designed to absorb shifts in model pricing and efficiency over time.

What Is Not Changing

Code completions and next edit suggestions remain unlimited on all paid plans. If your Copilot usage is primarily inline completion as you type, this billing change will barely register.

The shift matters most for chat, agentic sessions, and code review.

The Part That Has Developers Worried

The GitHub Community discussion thread on this change collected more than 400 comments and nearly 900 downvotes, which is unusually direct feedback for a GitHub announcement. The concern is not hard to understand.

One developer estimated that a single agentic coding session can consume $30 to $40 in credits. A Copilot Pro subscriber gets $10 per month. That is a mismatch of three or four times in a single session. Pro+ subscribers get $39, which fares better, but heavy agentic use can still exhaust that quickly.

The previous system had a fallback: when you ran out of premium requests, Copilot would drop back to a lower-cost model and keep working. That option is gone. Once you exhaust your included credits, Copilot stops unless you have enabled additional spending.

Budget Controls: The Safety Net

The good news is that GitHub has built explicit controls around spending. Admins can set budgets at four levels: per user, per cost centre, per organisation, and across the whole enterprise. If you leave your additional-spend limit at $0, Copilot simply stops when included credits run out. You cannot be charged anything beyond your subscription fee without explicitly enabling it.

For individual users, that means you can treat Copilot Pro as a strict $10/month tool if you want. You just need to be aware that intensive agentic work will hit that ceiling faster than it used to.

For organisation admins, the controls are more granular. You can cap individual users, ring-fence teams, or set enterprise-wide limits. Credits are also pooled across organisations in Business and Enterprise plans, so unused credits from lighter users can offset heavier usage elsewhere.

Annual Plan Subscribers: A Nuance Worth Knowing

If you are on an annual Copilot Pro or Pro+ plan, you are not automatically migrated on June 1. You stay on premium request-based billing until your plan renews. At renewal, you transition to Copilot Free, with the option to move to a monthly paid plan.

One catch: model multipliers for some models changed on June 1 regardless of your billing model. Annual plan subscribers may see different token rates even while still technically on PRU billing.

Code Review Now Uses Two Meters

Copilot code review has a dual billing structure that is worth flagging. Token consumption is billed in AI Credits (charged to the person who requests the review). The agentic infrastructure that runs the review also consumes GitHub Actions minutes, which are charged to the repository. Public repositories are unaffected since Actions minutes remain free for them.

The Auto-Mode Discount

One small offset: if you use auto model selection in Copilot Chat, Copilot CLI, or the cloud agent, you get a 10% discount on model costs while auto mode is active. GitHub picks the model; you pay slightly less for it.

Who This Affects Most

Light users and completions-focused developers will notice very little change. The unlimited completions policy covers the bulk of their usage.

Agentic and power users are the ones who need to think carefully. If you regularly run Copilot on multi-step tasks across a repository, your usage under the new model may be significantly higher than your included credits.

Organisation and enterprise admins have a new set of controls to configure, and a three-month promotional period (June 1 to September 1, 2026) where Business and Enterprise plans receive higher included credits than standard. That window is worth using to understand your actual usage patterns before the standard amounts kick in.

New subscribers will need to wait slightly longer: sign-ups for Copilot Student, Pro, Pro+, and Max remain paused for now, with GitHub indicating they plan to reopen in the coming weeks.

The Honest Take

This is a rational move for GitHub given the economics of agentic AI, and it mirrors how OpenAI, Anthropic, and Google price their APIs. The difference is that developers using Copilot are not API customers who expect to see token counts; they are subscription users who expected predictable monthly costs.

The included credits are reasonably generous for conversational chat use. They are not generous for agentic sessions, and that is where the frustration is coming from. The budget controls mean you are protected from surprise bills, but they also mean that agentic workflows will either cost more or hit a wall more quickly than before.

If you are a heavy Copilot user, now is the time to look at your actual session patterns, check whether Copilot Pro+ or Copilot Max better matches your usage, and set your spending limits deliberately rather than leaving them at defaults. The full pricing breakdown by model is available in GitHub’s documentation.