Copilot code review cost: AI credits plus Actions minutes

Since June 1, 2026, every Copilot code review on a private repository is billed twice: once in GitHub AI Credits for the model work, and once in GitHub Actions minutes for the runner the review agent runs on (changelog). On public repositories the minutes stay free. The change covers Copilot Pro, Pro+, Business and Enterprise, including reviews for organization members without a Copilot license (announcement).

Below: how each charge works, how to measure what your reviews use, which runners code review can run on, and how a flat per-review price compares.

Why a code review uses Actions minutes

Copilot code review runs its agentic features, such as gathering context from the whole repository and handing suggestions to Copilot cloud agent, as a GitHub Actions job. By default that job runs on a standard GitHub-hosted Ubuntu runner (code review docs). Before June 1 the runner time was free and reviews drew only on Copilot premium requests.

The two charges go to different places (models and pricing):

The AI credit part

GitHub doesn't disclose which model code review uses, and it picks the model automatically, so you can't work out a review's cost from the per-token price table. What GitHub does publish is an estimate per review (estimated consumption):

• Lite effort (the default): $0.05 to $1 in AI credits
• Balanced effort: $0.25 to $5 in AI credits

Cost goes up with pull request size and with repository custom instructions. Those ranges exclude Actions minutes. Balanced sends the PR to a higher-reasoning model, uses more credits, and "may consume marginally more GitHub Actions minutes" (review effort level).

On Copilot Business and Enterprise, each seat adds 1,900 or 3,900 AI credits a month (that's $19 or $39) to a pool shared across the organization (usage-based billing for organizations). Reviews draw from that pool along with Chat, the CLI and cloud agent. Once it's empty, reviews are billed as additional usage, which is on by default. Reviews for users without a Copilot license skip the pool entirely and bill straight to the organization as additional usage.

The Actions minutes part

A review's minutes come out of the same monthly allowance as your CI (GitHub Actions billing):

Minutes beyond the allowance cost $0.006 each on the standard Linux 2-core runner, and GitHub rounds every job up to the next whole minute (runner pricing). Larger runners never draw from included minutes, so each of their minutes is billed at their own rate.

GitHub doesn't publish how many minutes a typical review takes, so measure your own:

• In Actions metrics, filter by the copilot-pull-request-reviewer workflow (Actions metrics).
• In the billing usage report, filter workflow_path by dynamic/agents/copilot-pull-request-reviewer (billing reports).

Then the math is:

Review minutes per month = reviews per month × average billed minutes per review
Actions cost             = max(0, review minutes + CI minutes − included minutes) × $0.006

For a sense of scale, take 2,000 reviews a month at an assumed 4 billed minutes each. That's 8,000 minutes, more than twice GitHub Team's 3,000 included minutes, and $48 if all of it lands past the allowance. At GitHub's Lite estimate, the AI credits for those same reviews come to $100 to $2,000. The minutes use up included allowance that CI would otherwise have, but AI credits make up most of the dollar cost.

Which runner code review uses

A runs-on value in .github/workflows/copilot-code-review.yml sets the runner for Copilot code review. If that file doesn't exist, code review falls back to copilot-setup-steps.yml (configure runners). That second file is the one that configures Copilot cloud agent's environment, so without a copilot-code-review.yml, a runner change you make for the cloud agent changes code review's runner too.

GitHub documents three options:

• Standard GitHub-hosted runner. The default. Private-repo minutes count against your allowance.
• Larger GitHub-hosted runners. Set runs-on to the larger runner's label. You get more CPU, memory and disk, and every minute is billed at the larger runner's rate.
• Self-hosted runners through Actions Runner Controller (ARC). Set runs-on to your ARC scale set name. Self-hosted runners don't consume GitHub Actions minutes, but you run the Kubernetes cluster.

# .github/workflows/copilot-code-review.yml
jobs:
  copilot-setup-steps:
    runs-on: ubuntu-4-core # a larger runner label or an ARC scale set name

For self-hosted runners the docs are strict. They say "ARC is the only officially supported solution for self-hosting Copilot code review. For security reasons, do not use non-ARC self-hosted runners." Code review only runs on Ubuntu x64 Linux runners, and the runner's firewall has to allow api.githubcopilot.com, uploads.github.com and user-images.githubusercontent.com in addition to the standard Actions hosts.

Organization owners can set a default runner for code review and cloud agent across all repositories, either the standard GitHub runner or a labeled runner, and can stop repositories from overriding it (organization-level runners).

If an organization disables GitHub-hosted runners and has no self-hosted ones set up, reviews still run, but without the agentic features, and fall back to "a more limited review".

Controlling the bill

Most of the settings that decide the cost are in GitHub's code review configuration (configure code review):

• Review effort: Lite is the default. Making Balanced the repository or organization default moves GitHub's per-review estimate from $0.05 to $1 up to $0.25 to $5.
• Review new pushes: with this on, every push to an open PR gets a new review, and each one bills both meters. Without it, Copilot reviews a PR once, and you can ask for a re-review by hand.
• Review draft pull requests: drafts get reviewed before they're ready, so a PR can collect more reviews on its way to merge.
• Budgets: Actions budgets cap the minutes spend. User-level AI credit budgets, generally available since June 1, cap how much each person draws from the pool, and a user who hits theirs is blocked from code review along with other Copilot features.

Reviews and CI share one pool of included minutes. If CI is what pushes you past the allowance, moving CI jobs elsewhere frees minutes for reviews. GitHub Actions cost optimization covers that, including Tenki Runners at $0.002 per core-minute on x64. None of that moves the review job itself, since GitHub supports only its own runners and ARC for code review.

A per-review alternative: Tenki Code Reviewer

Tenki Code Reviewer is a GitHub App. There's no Actions workflow involved, so it uses no Actions minutes. A review costs $1.00 from workspace credits whatever the PR size (pricing). Reviews run when a PR is opened or updated, or when someone mentions @tenki-reviewer in a PR comment, and they follow the repository's AGENTS.md or CLAUDE.md (custom context). The Starter plan is free and comes with $100 in credits at signup, which is 100 reviews. Team is $250 a month ($200 billed yearly) and includes $100 in credits each month.

Per review, using GitHub's own estimates:

Which one costs less depends on where your reviews fall in GitHub's ranges, and GitHub doesn't say where a typical review falls:

• On Lite, Copilot usually costs less. GitHub's estimate tops out at $1 in credits per review, and reviews covered by seat credits you'd otherwise leave unused cost you only the minutes, or nothing within your included minutes.
• Tenki costs less on reviews that use more than $1 in credits plus minutes. By GitHub's estimate that means Balanced reviews, especially on large PRs, where the range goes up to $5.
• Tenki's bill is easier to predict, because the number of reviews is the whole cost.

Once you have a month of real usage, divide what code review cost you in credits and minutes by the number of reviews, and compare that against $1.00.

On quality, our code review benchmark (last updated May 2026) runs each tool at default settings on 50 pull requests that reintroduce 122 real bugs from open-source projects. Tenki caught 84 (68.9% recall), and Copilot caught 30 (24.6%). Tenki also posts the most comments, and its precision is about 0.30. It's our own benchmark, so read the methodology before relying on it. To try it, the quickstart takes you through installing the GitHub App.