GitHub Copilot Code Review Effort Levels Are GA: Lite vs Balanced

GitHub made the effort-level controls for Copilot code review generally available on August 7, 2026. The public preview levels Low and Medium are now the GA levels Lite and Balanced, and the feature ships with per-review selection, organization-level defaults, and visible labeling of which level actually ran on each pull request.

For teams running Copilot reviews at scale, this is the first real lever for controlling the cost and depth of AI review instead of treating every PR the same. Here is what changed, what each level actually costs, and how to configure it.

What shipped#

Copilot code review now supports two review effort levels on every plan that includes code review (Pro, Pro+, Max, Business, and Enterprise):

• Lite: the standard review. Fast, targeted feedback on common issues such as bugs, security vulnerabilities, and style inconsistencies. This is the default.
• Balanced: routes the review to a higher-reasoning model for longer analysis of complex logic, security-sensitive code, and cross-service changes.

Three behaviors make this usable in practice:

1. Per-review override: when you request a review, you pick a level for that review only. It does not change the repository or organization default.
2. Organization-level defaults: org admins set a default under Organization settings -> Copilot -> Copilot code review, and repositories inherit it unless they configure their own. Repository admins can override the org default for a specific repo.
3. Visible labels: the pull request overview comment and timeline events now state which effort level ran, so review depth is auditable across repositories.

Existing Low and Medium configurations carry forward automatically under the new names, so the rename does not break anyone who configured effort levels during the preview.

What Balanced actually costs#

The tradeoff is spelled out in the docs: Balanced reviews use more AI credits and more GitHub Actions minutes than Lite. The reasons are structural:

• Code review consumes AI credits for the model interaction (the review itself).
• The agentic capabilities, full project context gathering and passing suggestions to the Copilot cloud agent, run on GitHub Actions, so every review also consumes Actions minutes.

A Balanced review means a higher-reasoning model spends longer on the whole repository context, not just the diff. GitHub recommends larger or self-hosted runners for Balanced reviews, and larger GitHub-hosted runners bill at a higher per-minute rate. Self-hosted runners do not consume Actions minutes.

This is the same shape as the rest of the agent cost story: the bill is driven by model tier, context size, and runner class. For a team that auto-reviews every PR, switching the default from Lite to Balanced can multiply spend on routine changes without proportionally better feedback. The GA of effort levels is really the GA of a cost control for AI review.

How to use it#

Concrete defaults that map review depth to PR risk:

• Lite as the org default: routine changes, docs, small fixes, dependency bumps. Fast feedback matters more than exhaustive analysis.
• Balanced per-review: security-sensitive code, multi-service changes, migrations, anything touching auth or payments. Pick Balanced when you request that specific review.
• Repo-level exceptions: for repositories with strict quality standards, set Balanced as the repo default while keeping Lite everywhere else.

Where it fits#

Copilot code review is increasingly a full agent product rather than a diff scanner. It already gathers full project context, can hand suggested fixes to the Copilot cloud agent as a new PR, and can pull in repository agent skills and MCP servers during review. The effort levels sit on top of that: a way to say how deep the agent should go, per PR, without changing the tooling.

Two adjacent changes landed in the same window. GitHub also shipped MCP allowlists in enterprise managed settings on August 6, which gives admins control over which MCP servers Copilot can touch, and the Copilot impact dashboard added a return on investment section. The theme is consistent: GitHub is spending the summer on the governance layer around agentic review, not just on raw model quality.

The honest caveat still applies: Copilot is not guaranteed to catch everything, and GitHub says to validate its feedback alongside human review. Effort levels tune depth and cost; they do not replace judgment about what a change touches.

Continue Reading#

• The 12 best AI code review tools in 2026 - where Copilot review sits next to the alternatives
• PR governance for Copilot review - policies and settings for review at team scale
• A 400 dollar overnight agent bill and what it teaches - the cost mechanics behind agentic features
• Enterprise team model policy targeting in Copilot - controlling which models your org uses
• Switching from Copilot to Claude Code - what you give up and what you gain

Sources#

• Copilot code review effort levels are generally available - GitHub Changelog, August 7, 2026
• About GitHub Copilot code review - GitHub Docs
• GitHub Copilot weekly releases - August 3 - GitHub Changelog