Copilot Code Review Limits Bite After Balanced Default Switch

What changed

GitHub's Copilot code review has become more expensive to run, and at least one automation-heavy maintainer is already hitting a ceiling.

On September 28, 2026, GitHub changed the Default review effort level for Copilot code review from Lite to Balanced. The change applies to both new and existing repositories and organizations, following an announcement on August 28. 3 Anyone who had explicitly chosen Lite kept that setting. Everyone who had left the option on Default received the heavier review level automatically. 3

In the same October 2 update, GitHub added REST and GraphQL API support for requesting Copilot reviews. Callers can set an effort level on each request, which lets teams start reviews from their own scripts, workflows, and internal tools. 3 The features are generally available on the Pro, Pro+, Max, Business, and Enterprise plans. 3

The weekly limit nobody budgeted for

A day before that post, maintainer ptr727 opened an issue in the ProjectTemplate repository. It reported that Copilot code review now reaches a weekly rate limit under the project fleet's normal review traffic. 2 The issue sets out several observations.

• The weekly limit is separate from the monthly budget. A new billing month does not reset it. Once the limit is reached, every review request fails until the stated reset time. 2
• The limit itself is not new. The issue says hitting it is new, and attributes that to the default level moving from "lite" to "medium." Each review now uses more capacity. 2 The issue's "medium" appears to match what GitHub calls Balanced.
• The available controls do not match the error message. According to the issue, settings let users change the review level but not the model. The failure message tells users to switch to an auto model, yet the maintainer found no setting that offers this. 2
• The failure is hard to see. The issue says it is invisible to the fleet's tooling. 2 For a fleet that relies on automated review, that means reviews can stop without anyone noticing.

The cost backdrop

This lands on top of an earlier pricing change. GitHub's documentation on premium requests says that from June 1, 2026, Copilot code review carries a model multiplier of 13. Each pull request review, or review of code in the IDE, uses 13 premium requests from the monthly quota. 1 By comparison, Copilot Chat uses one premium request per prompt multiplied by the model's rate. Copilot CLI uses one per prompt on the default model. 1

So two separate pressures now affect the same feature:

1. A monthly premium-request quota. Code review already drew heavily on it after June. 1
2. A weekly rate limit. This is not mentioned in the excerpted documentation or the announcement, but it is now being hit because Balanced reviews cost more. 23

GitHub's announcement presents Balanced as a better default and tells users who prefer Lite that they can switch back at whatever level they manage. 3 It does not mention how the change interacts with weekly limits. The maintainer's account suggests that, for heavy users, this interaction is the most important practical effect. 2

Where the accounts diverge

GitHub's framing is about choice: pick Lite, Balanced, or another level, now even per request through the API. 3 The issue describes a narrower set of options in practice. Only the level can be adjusted, the error message points to a setting that does not seem to exist, and failures are not surfaced. 2 Both accounts can be accurate at once. The controls exist, but they may not meet the needs of an automated fleet trying to stay under a limit it cannot easily see.

The terminology also differs. GitHub's docs and blog use "Lite" and "Balanced" for effort levels. 3 The issue uses "medium," and talks about an auto model. 2 That difference reflects the confusion the issue describes about which setting controls what.

Why it matters

The timing is awkward for GitHub. On the same day it made Copilot review callable from scripts and pipelines, a project already running Copilot review at fleet scale reported that the new default can lock it out for the rest of the week. 23

The per-request effort parameter in the new API is probably the most useful tool for teams in this position. 3 Automation could request Lite for routine changes and keep Balanced for pull requests that warrant closer review. In effect, review depth becomes something to allocate, which is the question the ProjectTemplate issue asks. 2

The broader point concerns defaults. Changing the default affected everyone who had never touched the setting, and it is the most hands-off users, often automated ones, who are least likely to notice. Until GitHub documents the weekly limit and gives clear signals when it is reached, teams relying on Copilot review in CI should do three things:

• Check their effort settings.
• Consider setting Lite explicitly where depth matters less.
• Treat a failed review request as an error to monitor, not something to ignore.