GitHub announced it quietly. Copilot code review — the feature that automatically reviews your pull requests — will start consuming GitHub Actions minutes on June 1, 2026.

Update, October 2026: it happened, and the bill got a second meter. Since June 1, 2026, Copilot bills in GitHub AI Credits (1 credit = $0.01) for the input, output and cached tokens each request uses, instead of premium requests, and every code review is billed both ways: AI credits for the tokens, Actions minutes for the runner on private repositories (GitHub, April 2026; models and pricing). Plan prices stayed the same, and each now includes that amount in monthly credits: Pro $10, Pro+ $39, Business $19 per user, Enterprise $39 per user. The rates and the cost steps below are updated to match.

If you're on a free plan, you get 2,000 Actions minutes per month, and every Copilot review of a private repository draws from them. On paid plans, the math is worse. You're already paying $19 per user for Copilot Business. Now you're paying again for the compute to run it.

This isn't a price increase. It's a pricing layer cake.

What Actually Changes on June 1

GitHub's changelog, published April 27, says:

"GitHub Actions minutes will be consumed from your existing plan entitlement for each review that is run on private repositories, with any usage beyond your included minutes billed at standard GitHub Actions rates."

Translation: every Copilot code review on a private repo now counts against your Actions budget. Public repositories are exempt, which is nice if you work exclusively in open source. Most companies don't.

The billing works like this:

  • Free plans: 2,000 minutes/month. Copilot reviews eat into this directly.
  • Team plans: 3,000 minutes/month included, then $0.006 per minute on the standard Linux runner that code review uses by default.
  • Enterprise plans: 50,000 minutes/month included, then same overage rate.

GitHub doesn't publish how many minutes a typical review takes. Copilot code review runs on a standard GitHub-hosted Ubuntu runner by default, so the time depends on your pull request, and you can measure it (step 1 below) (GitHub Actions billing docs). A self-hosted or larger runner bills at a different rate, and since June organization admins can set one as the default for every repository (GitHub changelog, June 2026).

Reviews run when someone requests one, or automatically where a branch ruleset or an organization setting turns automatic review on. One ruleset option matters for the bill: with Review new pushes on, Copilot reviews every push to the pull request; leave it off and it reviews the pull request once (GitHub Docs).

Why GitHub Is Doing This

GitHub's stated reason is that usage-based billing "aligns Copilot pricing with actual usage" (GitHub, April 2026). Copilot code review runs inference on GitHub's infrastructure. Inference costs money. GitHub wants you to pay for it rather than subsidizing it from Copilot subscription revenue.

That's fair, in isolation. But the timing isn't innocent.

Copilot Business costs $19 per user per month, and since June that price is also the monthly AI-credit allowance: once an organization's pooled credits run out, it either pays per token at published rates or hits the spending cap it set. And now GitHub's adding a variable compute cost on top of the fixed subscription.

This is classic vendor lock-in arithmetic. Once your team is trained on Copilot workflows — code review, chat, completions — switching costs are high. You've got muscle memory, prompt libraries, and institutional knowledge. GitHub knows it, so it can raise prices incrementally without losing customers right away.

First came premium requests with per-model multipliers. Then Actions minutes for code review. Then, on June 1, per-token AI credits, with the old fallback to a cheaper model gone: "Fallback experiences will no longer be available," GitHub wrote. Each step's small enough that migration feels like more work than just paying.

The Real Cost Is Not Money

The dollar cost's annoying. The structural cost's worse.

When your code review infrastructure is tied to GitHub's billing model, GitHub controls your engineering process. They decide what features cost extra. They decide when to throttle. They decide which models power the review. You don't. Anthropic is playing the same game with Claude rate limits — this is an industry pattern, not a GitHub quirk.

This month it's Actions minutes. Next month it could be storage for Copilot chat history. Then API access for custom integrations. Each new charge is a new dependency. Each dependency makes migration harder.

Migration is already hard. GitHub owns your repositories, your issues, your CI/CD, your project management, and now your AI code review. That's not a tool stack. That's a platform prison.

What Open Source Looks Like Instead

The alternative isn't "find a cheaper GitHub." The alternative's decouple your AI tooling from your git hosting.

Octomind is an open-source AI agent that runs locally or on your own infrastructure. It reviews code, refactors modules, writes tests, and debugs errors. It doesn't bill per minute. It doesn't require a subscription to a platform that also hosts your code. There's even a GitHub Action that posts a PR brief on every pull request — the same job as Copilot code review, with the model and the token bill in your hands. The Action itself still runs on a runner, so on private repositories it uses your Actions minutes unless you point it at a self-hosted runner.

In practice, that means:

  • No platform meter: Run as many reviews as your hardware, your provider's rate limits, or your API budget allow.
  • No platform lock-in: Octomind works with any git provider — GitHub, GitLab, Gitea, or bare git. Switch hosts without changing your AI workflow.
  • Model choice: Use Claude, GPT, DeepSeek, or a local model via Ollama. You pick the capability-to-price ratio that fits your team.
  • Transparent costs: API costs are linear and predictable. No surprise overage fees. No bundled pricing that hides true costs.

The tradeoff is setup. GitHub Copilot works out of the box because GitHub controls the entire stack. Octomind needs configuration: install it, sign in with octomind login or bring your own API keys, pick a model, and tune the review prompt for your codebase. You do that once.

How to Estimate Your Copilot Code Review Cost

If you're staying on Copilot, budget for it this way.

Step 1: Measure what reviews already use. In your organization's GitHub Actions metrics, filter by the copilot-pull-request-reviewer workflow; in the billing usage report, filter workflow_path by dynamic/agents/copilot-pull-request-reviewer (GitHub Docs). That gives you the number of review runs and their minutes for the last 30 days.

Step 2: Take the average minutes per review from those runs. GitHub doesn't publish a typical figure, and it varies with the size of the change.

Step 3: Multiply reviews by minutes, then by your overage rate. On the standard Linux runner, overage is $0.006 per minute. 50 reviews at 20 minutes each = 1,000 minutes. If your Team plan includes 3,000 minutes and you use 2,000 for CI already, that is 1,000 minutes of overage = $6 per month. Small, but it scales with team size.

Step 4: Add the AI credits. Code review picks its model automatically and doesn't disclose it, so the per-token cost varies between reviews; read the actual credits from the billing usage report. They're charged to whoever requested the review, or to the pull request's author when a policy triggered it (GitHub Docs).

Step 5: Add the mental tax. Someone on your team now has to monitor Copilot usage, argue about whether a review is worth the minutes and credits, and maybe turn off automatic review on certain repositories. That's not engineering work. That's vendor management.

The Bigger Picture

This Copilot code review pricing change's a signal. GitHub is moving from "AI features included" to "AI features metered." That's the correct business model for compute-intensive services. It's also the model that makes platform lock-in profitable.

When every AI feature's got a meter, the platform that owns your data can set the meters however they want. They can bundle, unbundle, discount, and surcharge in ways that make comparison shopping impossible. Is $19 Copilot plus $6 in Actions minutes plus the credits each review draws cheaper than paying for tokens directly? Depends on your usage. Depends on your repo sizes. Depends on GitHub's mood.

The only way to win this game is not to play it.

Run your AI tooling independently. Pay for compute directly. Own your model choices. When GitHub raises prices, you shrug. When they change terms, you don't care. Your AI workflow stays yours.

Octomind is built for this. Open source. Multi-provider. Local-first. No meters except the ones you set.

Code review shouldn't be a billing surprise.

Run Octomind locally → github.com/muvon/octomind
Set up Ollama for free local models → ollama.com