← All posts

FinOpsAid blog

Copilot code review now has a cost dial. Someone turned it up on September 28.

If nobody in your organization has opened the Copilot code review settings, your reviews got more thorough on September 28. They probably also got more expensive. GitHub didn't publish by how much, and nothing on your invoice will say why.

That was the first of four changes in five weeks. Read together, they make review effort a cost setting with three positions, an admin ceiling and a choice of who pays. Here's what each one does and which defaults are worth replacing.

What changed between September 28 and October 28

Four announcements, all for Copilot Pro, Pro+, Max, Business and Enterprise:

Date Change Default
Sep 28 Default review effort moved to Balanced for new and existing repos and orgs Balanced, unless you'd explicitly chosen Lite
Oct 2 Reviews can be requested through the REST and GraphQL APIs, with effort set per request Not used unless someone scripts it
Oct 8 Choose how members with a Copilot license are billed: Member or Organization Member
Oct 8 Only allow Copilot code review to be triggered by authorized users Off
Oct 28 Max effort level, public preview Not selected; admins can cap the maximum

Sources: GitHub Changelog, API support and new default effort level (October 2), new organization billing options and controls (October 8), and Max and approvals (October 28).

Approvals also went GA on October 28. We covered the approvals side in September, including the fact that a review bills both AI credits and Actions minutes, in Copilot can approve pull requests now. This post is about the effort and billing settings around it.

The default that moved without you

GitHub's wording is precise: the Default review effort level "now uses Balanced for new and existing repositories and organizations". If you had explicitly selected Lite, that choice was kept. If you'd never touched the setting, you were moved up.

That's a fair product decision. Balanced presumably catches more, and most teams never pick a setting. But it's also a spend change made on your behalf, and GitHub published no credit figure for Lite versus Balanced. So you can't price the move from the changelog. You can only see it in consumption afterwards.

Settings cascade from enterprise to organization to repository to personal, and each level can override the one above. So "did we pick Lite?" doesn't have one answer for most enterprises. It has one answer per level where someone did or didn't make a choice.

Effort is now a three-position cost dial

Max adds a third position, and it's honest about what it costs. Max "runs several independent review agents across models from different providers", and then an aggregation agent reconciles their findings. GitHub says plainly that Max "takes longer and uses more AI credits than Lite or Balanced".

No multiple is published. We won't guess one. Several agents plus an aggregator is structurally more work than one reviewer, so the direction is clear even though the size isn't.

The important control is the ceiling. Enterprise, organization and repository admins can set a maximum review effort level, and anything above it isn't available below that level. Max can be picked per pull request from the Reviewers bar, or set as a default at any level. So without a ceiling, anyone who can request a review can choose the most expensive option on any PR.

Our analysis: set the ceiling while Max is still in preview, not after it ships GA and becomes familiar. A ceiling of Balanced everywhere, raised to Max only on the repositories where GitHub itself suggests it (auth logic, shared libraries), is a policy you can explain. Max everywhere by accident isn't.

Who pays decides what fails

The October 8 billing setting matters more than it looks. It has two options:

  • Member (default). The review bills the requesting member's own Copilot entitlement. GitHub's words: "If the member's quota is exhausted, the code review fails."
  • Organization. The review bills the organization that owns the repository. This needs AI Credits paid usage enabled, and you can optionally set a budget.

Read the default from a release manager's point of view. Your heaviest Copilot users are often your most active reviewers. Under Member billing, when one of them runs out of quota mid-month, their review requests start failing. If your repository counts Copilot's approval toward required approvals, a failed review isn't a cost event. It's a merge that can't happen.

What happens when credits run out, by review effort level and billing setting A grid with three columns for Copilot code review effort levels, Lite, Balanced and Max, and two rows for the billing setting, Member and Organization. Effort rises from left to right: Lite is the lightest, Balanced has been the default since September 28, 2026, and Max runs several review agents plus an aggregation agent and uses more AI credits, with no published multiple. Under Member billing, the default, each review draws on the requesting member's own Copilot entitlement, and when that quota is exhausted the review fails. Higher effort reaches that point sooner. Under Organization billing, the review draws on the organization's AI credits with paid usage enabled, and an optional budget set by the organization is the control. Admins can set a maximum effort level so that Max is unavailable where it is not wanted. The figure is conceptual and contains no consumption numbers. WHEN CREDITS RUN OUT — EFFORT × WHO PAYS

Lite Balanced Max default since Sep 28 several agents + aggregator more AI credits per review →

Member the default Draws on the requesting member's own entitlement. Quota exhausted → the review fails. Higher effort gets there sooner. Heaviest users hit it first. Organization opt-in Draws on the repo-owning org, with AI Credits paid usage on. The control is a budget the organization sets. Members' own quotas are left for chat and agents. A maximum effort level, set by admins, decides whether the Max column exists for you at all.
Conceptual, no consumption figures. GitHub hasn't published credits per review at any effort level. Billing behaviour and the failure on quota exhaustion are from the October 8, 2026 changelog; Max's design and the maximum-effort control are from the October 28, 2026 changelog. Both retrieved 2026-11-03.

Organization billing moves the question from "whose quota ran out?" to "what budget did we set?" That's the more governable of the two. It also keeps reviews from draining a developer's allowance for the chat and agent work they'd rather spend it on. The cost lands on the organization that owns the repository, which is usually where it belongs anyway. For how that spend rolls up, see cost centers and AI credit pools.

The companion control is the authorized-users setting. Turn it on and reviews have to come from a Copilot license your organization or enterprise provides. Personal or other external licenses can't trigger them. Set at org level, repository admins can't turn it off. It's a governance control first, but it also settles which licenses review spend can come from.

Reviews by API are spend by script

The October 2 change is small in the changelog and large in practice. Anyone with access can now request a Copilot review from a script, a workflow or an internal tool, and choose the effort level per request.

That's useful. It's also the same shape as an unattended automation: something other than a person decides when credits get spent, and at what effort. A CI step that requests a review on every push at Max is a perfectly valid API call.

Our analysis: treat API-requested reviews the way you'd treat Copilot Automations. Know which systems call it, what effort they pass, and which billing setting they land on. GitHub says levels above the maximum-effort ceiling "will not be available to use within the enterprise, organization, or repository". The changelog doesn't say in so many words that this covers API requests, so test it in your own tenant before relying on it. Set the ceiling anyway, before anyone writes the script.

Our finding: if the September 28 default bump changed your spend, you won't see it labelled anywhere, including in FinOpsAid. The per-user AI credit figure GitHub exposes, which is what we read, is one daily number per person with no feature breakdown. So review credits arrive mixed in with chat and agent credits, and a change in review effort shows up only as an unexplained step in a few users' daily totals. Our suggestion, not something we've measured: compare per-user credit trends for the two weeks before September 28 with the two weeks after, and look at your most active reviewers first. A step change there, with no matching change in chat use, is the effort default showing through.

What to set this month

  1. Decide the default effort level on purpose. If Balanced is right, leave it, but choose it. If you want Lite, set it at the highest level you manage so every org and repo below inherits it.
  2. Set a maximum effort level now. Balanced as the ceiling, with Max allowed only on the repositories where a missed issue costs the most.
  3. Switch billing to Organization where Copilot approvals count. On those repositories, a review that fails on someone's quota blocks a merge. Pair the switch with an organization budget.
  4. Turn on authorized-users-only at org level if you want review spend to come only from licenses you provide.
  5. List what calls the review API. Scripts, CI steps, internal bots. Note the effort each one passes.
  6. Check the timing against your billing model. If you move to GitHub's new Copilot access model, where usage is metered from the first request with no included allowance, these settings decide your code review line from day one. See what the new model costs.

Where FinOpsAid fits

FinOpsAid reads your GitHub org read-only, syncs nightly, and shows AI credit consumption per user, per team and per cost center next to the rest of the GitHub bill. That's enough to spot a step change after a settings change, and to see which reviewers carry the most consumption. The limits are worth stating plainly. GitHub's per-user AI credit figure is one daily number with no feature breakdown, so we can't split code review credits from chat credits. GitHub's Copilot metrics lag roughly two to three days. For a hard stop, use GitHub's own budgets (how user budgets enforce). Related: what actually consumes GitHub AI credits and how to read the Copilot page.

Frequently asked questions

What is the default Copilot code review effort level?

Balanced, since September 28, 2026. GitHub moved the Default setting from Lite to Balanced for new and existing repositories and organizations. If you'd explicitly selected Lite, that choice was kept. You can change it at enterprise, organization, repository or personal level, and each level overrides the one above it.

Does Copilot code review Max cost more?

Yes. GitHub says Max takes longer and uses more AI credits than Lite or Balanced, because it runs several independent review agents across models from different providers and then an aggregation agent. GitHub hasn't published how many more credits. Admins can set a maximum effort level to keep Max unavailable where it isn't wanted.

Why did a Copilot code review fail with a quota error?

Under the default Member billing setting, a review bills the requesting member's own Copilot entitlement, and GitHub says the review fails if that quota is exhausted. Organization owners can switch to Organization billing in Copilot policies, which bills the repository's organization instead. That requires AI Credits paid usage and can carry an optional budget.

Sources

If you want to see whether September 28 moved your most active reviewers' credit consumption, connect your GitHub org — read-only and free during beta — or explore the demo dashboard first.

copilotai-creditscode-reviewbillinggovernancefinops

See it in your own org

FinOpsAid meters the GitHub spend no invoice itemizes — read-only, and free during beta.