If you use GitHub Copilot, you probably already got the communication that there's a billing change coming on June 1, 2026 that you should understand before it kicks in. GitHub is moving from premium request units (PRUs) to a token-based credit system, and for heavy users, especially those doing agentic work, the cost difference can be significant. The good news (I don't really know if I should call it that): GitHub has shipped a billing preview tool that lets you see your projected costs right now.
Here's what's changing, what the impact is, and exactly how to use the preview experience to protect your budget.
What's actually changing?
GitHub Copilot has historically billed "premium requests" — a flat unit that counted each interaction with advanced models. Starting June 1, that system goes away. In its place is GitHub AI Credits, priced by token consumption: every input token, output token, and cached token is metered at the published API rate for whichever model you're using.
The key things to know:
- Base subscription prices are not changing. Copilot Pro stays at $10/month, Pro+ at $39/month, Business at $19/user/month, and Enterprise at $39/user/month.
- Each plan includes a matching credit allotment. A $19/month Business seat comes with $19 in monthly AI Credits. Once you exhaust those credits, you can buy more or stop.
- Code completions and Next Edit Suggestions remain unlimited and do not consume AI Credits. The changes apply to Copilot Chat, the cloud agent, Copilot Spaces, Copilot CLI, Spark, and third-party coding agents.
- Code review now also consumes GitHub Actions minutes. Copilot's code review feature moved to an agentic architecture running on GitHub Actions. From June 1, each review on a private repo will draw from your included Actions minutes, with overages billed at standard Actions rates.
- Fallback to cheaper models is going away. Today, if you exhaust your PRUs, Copilot falls back to a less capable model. Under the new system, usage is governed purely by your available credits and any admin budget controls you set.
For monthly Pro and Pro+ subscribers, the migration happens automatically on June 1. Annual plan subscribers stay on their existing plan (with updated model multipliers from June 1) until their plan expires, at which point they convert to Copilot Free or upgrade to a monthly plan.
Why is this happening?
GitHub is straightforward about the reason: Copilot is a much more compute-intensive product than it was a year ago. Agentic workflows; agents that loop through tool calls, pull in wide repository context, and orchestrate multi-step tasks; consume far more infrastructure than a simple autocomplete suggestion. The PRU system was designed for a simpler era, and it no longer reflects actual cost of service.
I always reminded my team that GitHub Copilot was priced too low. So this change isn’t unexpected.
The shift mirrors what other AI coding tools have done. Cursor moved to a credit-pool model in mid-2025. Anthropic charges Claude Code on a token basis. OpenAI moved Codex to token-based credits. Usage-based billing is becoming the industry standard for AI tools.
Who's most impacted?
If your Copilot usage is mostly code completions and occasional quick chat queries, you'll likely come out fine or even benefit from the change, since lightweight models are cheap per token.
If you regularly use:
- Frontier models (GPT-4.1, GPT-5, Claude Sonnet, Gemini) for long conversations
- Agentic sessions with the Copilot cloud agent or CLI in autopilot mode
- Copilot code review on busy repos with many pull requests
- Large context windows — long files, sprawling codebases sent as context
…then your token consumption is meaningfully higher, and your bill under the new model could increase substantially. Some early testers in the GitHub community have reported large differences between their PRU cost and projected AI Credit cost for the same April 2026 usage.
How to check your impact?
GitHub launched a billing preview experience to give users and admins visibility before the June 1 cutover. Here's how to use it.
For individual users (Copilot Pro / Pro+)
- Log in to github.com and go to your Billing Overview page.
- Navigate to the Premium Request Analytics page. You'll see an announcement banner — click "Preview your usage".
- In the dialog, click "Download CSV" to request your detailed usage report.
- The report is generated asynchronously and sent to your email.
- Once you have the CSV, upload it to the billing preview tool at copilot-billing-preview.github.com for a full breakdown.
The CSV includes your existing billing columns plus two new ones showing the estimated AI Credit equivalent for each interaction — so you can see model by model, day by day, where your usage lands.
For admins (Copilot Business / Enterprise)
- Log in and go to your enterprise home page, billing overview, or premium request analytics page.
- Click "Preview your usage" from the announcement banner.
- Download the CSV — it is prepared asynchronously and you are noticed by email. The downloaded file includes one row per user, per model, per day.
- Upload to the billing preview tool for a breakdown by user, model, and surface.
The report is also available via the API if you want to pull it programmatically. GitHub has also made April 2026 data available specifically to help you prepare — this is the most useful month to use since it reflects real recent usage patterns. Note: there are some known duplicate entries for April 24–30 due to a data backfill gap, so treat the numbers as directional rather than exact.
What to do next?
Once you've seen your projected numbers, there are a few levers worth pulling:
- Understand where your credits are going. Look at which models and surfaces are driving consumption. A single agentic session with a frontier model can use dramatically more tokens than dozens of Chat interactions with a lightweight model.
- Consider model choice. If you're reaching for GPT-5 or Claude Sonnet for questions where a lighter model would do, switching can meaningfully reduce token consumption. Copilot supports a range of models — use the right tool for the task.
- Set budget controls. Admins on Business and Enterprise plans can configure budget limits to prevent unexpected overages. For personal accounts, GitHub is rolling out equivalent controls. This is the safety net for agentic workflows, which are the hardest to predict in advance.
- Communicate with your team. If you're an admin, your developers may not know this change is coming. Let them know that agentic sessions and code review are metered differently from June 1, and point them to the preview tool so they can check their own usage.
- Consider plan timing. If you're on an annual Pro or Pro+ plan, you won't auto-migrate on June 1 — but your model multipliers will increase that day. You may want to evaluate whether converting to a monthly plan (with prorated credits for your remaining annual value) makes sense before your annual plan expires.
Conclusion
For some developers using Copilot for code completions and occasional chat, this change is largely neutral. The shift to AI Credits essentially means you're paying for what you use, and lightweight interactions remain cheap.
For teams going all-in on AI; running agentic workflows, deploying the cloud agent, running Copilot in autopilot mode, doing automated code review across many PR; the impact could be big.
The preview tool exists precisely so you're not surprised on June 1. Use it now, understand your baseline, and set budget controls before the transition lands.
I'm currently planning to assign different budgets depending on the usage type. We'll see how it turns out. To be continued...
More information
Preparing for your move to usage-based billing - GitHub Docs
Preparing your organization for usage-based billing - GitHub Docs