By On June 1, 2026 — five days from now — GitHub is retiring its Premium Request Unit system and switching all Copilot plans to token-metered AI Credits. The plan prices are not changing. GitHub will tell you that repeatedly. What they won’t lead with: the value you get for those prices is changing substantially for anyone who runs agents, automated code review, or multi-step workflows. Before the deadline hits, here’s what you need to understand.
From Flat Rate to Token Meter
Under the old Premium Request Unit (PRU) system, every premium Copilot interaction — whether a quick chat question or a 20-step agentic session pulling context from your entire codebase — cost a flat rate. Roughly $0.04 per request, regardless of how much compute it actually consumed. That was predictable. Developers could plan around it.
Starting June 1, you pay by tokens: input tokens + output tokens + cached tokens, at per-model rates, converted to GitHub AI Credits at $0.01 each. A short, focused chat message may cost less than a PRU. A long agentic session with an Opus-class model is a different story. One developer who ran the new Billing Preview tool against April’s usage — an extreme but real example — saw a $39 PRU estimate balloon to $902 under token billing. That’s the tail risk GitHub doesn’t advertise.
The credit math is straightforward once you accept what it implies. Copilot Pro ($10/month) includes 1,000 AI Credits. Pro+ ($39/month) includes 3,900. A single Opus 4.7 session with 10,000 input tokens and 2,000 output tokens costs roughly 30 credits. Pro plan users have enough for about 33 such sessions per month before hitting the ceiling. For completion-heavy developers, this is fine. For anyone running sustained agent loops, it’s a constraint that didn’t exist before.
Code Review Now Bills Twice
This is the change that’s getting less attention, and it should get more. Starting June 1, Copilot code review triggers two separate charges: AI Credits for the language model processing your pull request, and GitHub Actions minutes for the agentic infrastructure running on private repositories. These come from the same plans — your CI/CD budget and your Copilot credit pool both take hits simultaneously.
Teams running automated PR reviews at scale will want to audit their review frequency before June 1. This is not a corner case. Any team that enabled Copilot code review and runs multiple PRs daily is about to discover a new line item in their Actions bill.
What Stays Free
Code completions and Next Edit Suggestions do not consume AI Credits. They remain unlimited for all paid plans. For developers whose Copilot use is primarily inline completions — which is still the majority of users — the June 1 change may be nearly invisible. The pain is concentrated in chat, agents, Copilot Workspace, and code review. Know which category describes your workflow before you panic or dismiss the change.
The Concurrent Plan Cuts
The billing switch does not land in isolation. GitHub also paused new sign-ups for Pro, Pro+, and Student plans in May, removed all Gemini models from Copilot on the web, and stripped Opus 4.5 and 4.6 from Pro+, leaving only Opus 4.7 available there. The company’s stated reason: agentic usage has intensified to the point where “a handful of requests can exceed the plan price.” In plain terms, GitHub’s flat-rate model got exploited by power users, and the fix is being distributed across all customers.
The promotional credits buffer the landing for Business and Enterprise customers — $30/user instead of $19 for Business, $70/user instead of $39 for Enterprise, through August 2026. After that, it’s baseline. Set a strategy before the promotional window closes.
What to Do Before June 1
Five concrete steps before the meter starts:
- Run the Billing Preview: Access it via the premium request analytics page in GitHub settings. Download the CSV. See what your actual April usage would have cost under token billing. This takes ten minutes and eliminates guesswork.
- Audit code review usage: Check how many PRs your team reviews with Copilot per day on private repos. Multiply by what it now costs in both credits and Actions minutes.
- Trim your MCP tools: Each MCP tool definition costs 100–500 tokens per agent step. Disable servers your workflows do not actively need. GitHub’s own token efficiency guidance puts potential savings from this at 20% or more on tool-heavy sessions.
- Set spend limits: Admins on Business and Enterprise can cap spending at the enterprise, cost center, or user level. Do this before June 1, not after your first surprise bill.
- Know your model costs: Haiku-class models cost a fraction of Opus. For tasks that do not need frontier reasoning, dropping a model tier saves real money under token billing. This was not a consideration under PRUs.
The Competitive Reality
Cursor Pro ($20/month) and Windsurf Pro ($20/month) still use request-based billing — predictable, bounded costs regardless of context size. Copilot Pro ($10/month) remains the cheapest entry point in the market. But for developers running heavy agentic workflows, the new token model means Copilot could become the most expensive option despite its lower base price. That’s a real shift in the competitive calculus worth factoring into your toolchain decisions.
The flat-rate era of AI coding subscriptions is ending, and GitHub is leading the way. Token billing aligns cost with actual compute consumed — that’s a defensible model. But it puts the burden of cost awareness on developers in a way it never did before. Run the preview, audit your workflows, and set your limits. The meter starts in five days.
