Copilot code review's Balanced default may lift PR costs

GitHub has changed how much work Copilot code review does on a pull request when nobody picks a setting. Since September 28, 2026, the Default effort level runs the deeper Balanced mode instead of the lighter Lite mode 4. On October 2, GitHub also opened Copilot code review to its REST and GraphQL APIs 14.

Each change is small. Together, and on top of a billing model that now charges in two currencies, they matter to any team that pays for Copilot reviews on private repositories.

What changed

The October 2 update has two parts. First, developers can request a Copilot review through the supported REST and GraphQL APIs. Each request can set its own effort level 4. GitHub says this lets teams start reviews from their own scripts, workflows and internal tools 4.

Second, Balanced is now the default effort level for new and existing repositories and organizations that use Copilot code review 4. GitHub announced the switch on August 28, and it took effect a month later 4. Teams that had explicitly chosen Lite kept that choice 4. Teams that left the setting on Default were moved to Balanced.

The settings are layered, and each level can override the one above it 1. An organization can set a default that its repositories inherit while still controlling individual reviews 2. Both changes are generally available on Copilot Pro, Pro+, Max, Business and Enterprise plans 14.

Where the effort levels came from

Lite and Balanced came out of the public preview, where they were called Low and Medium 2. GitHub renamed them when they became generally available, and earlier settings carried over to the new names automatically 2.

GitHub describes the two tiers this way:

• Lite is for straightforward changes such as documentation updates and small fixes 2.
• Balanced is for changes that need deeper analysis "from a higher-reasoning model," such as complex logic, security-sensitive code and cross-service changes 2.

The premise is that not every pull request deserves the same scrutiny. Making the deeper tier the default sits awkwardly with that premise. The tool is now tuned for the harder cases, and teams have to opt down for routine ones.

The billing layer underneath

The cost question comes from how reviews are billed. Since June 1, 2026, every Copilot code review on a private repository has been charged twice [3]:

• GitHub AI Credits pay for the tokens the review model uses. They are charged to whoever requested the review, or to the PR author for automatic reviews. If that person has no Copilot seat, the organization pays 3.
• GitHub Actions minutes pay for the runner time of the review job. They come out of the repository owner's Actions allowance 3.

The Actions charge exists because the review's agentic features run as a GitHub Actions job, by default on a standard GitHub-hosted Ubuntu runner 3. Those features include gathering context from across the repository and passing suggestions to Copilot cloud agent. Before June 1 that runner time was free, and reviews drew only on Copilot premium requests 3. On public repositories the minutes are still free. The billing change also covers reviews for organization members who have no Copilot license 3.

Why the default matters for spend

None of GitHub's announcements say how much more a Balanced review costs than a Lite one. Any specific multiplier should be treated with caution. The direction is still easy to infer. A higher-reasoning model doing deeper analysis will usually consume more tokens, and token use is what AI Credits measure 23. A more thorough review may also run longer on the Actions runner, though this has not been published.

The main risk is inattention. Organizations that never touched the effort setting were moved to the more expensive tier without doing anything 4. Automatic reviews are billed to the PR author or, for unlicensed members, to the organization 3. A busy repository could therefore see higher spend that nobody chose.

API access adds to this. Once reviews can be triggered from scripts and internal tooling 4, review volume is set by automation rather than by people clicking a button. A pipeline that calls the API without setting an effort level will presumably fall back to the inherited default, which is now Balanced.

What teams should do

GitHub has not hidden any of this. The default change was announced a month ahead 4, the billing rules are published 3, and Lite can be selected at whichever level an admin manages 4.

The practical issue is that the cost of a review now depends on configuration choices many teams have never made. Teams that care about spend should:

1. Set the effort level explicitly at the organization level rather than relying on Default.
2. Reserve Balanced for repositories or pull requests where the risk justifies it.
3. Specify the effort level in every API call, so automation does not inherit the more expensive setting by accident.

The new controls are useful. They only save money if teams actually set them.