It depends on which Claude review ran: "Claude has already reviewed" can mean one of six things, five of which Anthropic lists, and /code-review reports in your session, posting to the pull request only with --comment. Copilot code review analyzes the entire repository for context and also reads CLAUDE.md, which /code-review and the managed Code Review read too, so some rules are shared. GitHub estimates an average review at $0.25 on Lite, $1.50 on Balanced (the default) and $10 on Max, excluding Actions minutes, and says actual costs can fall outside its estimates. GitHub's pages, as of 5 November 2026, describe models from different providers only for Max, which is "currently in public preview and subject to change." Your own pull requests settle the rest: count what only the second reviewer found and you applied.
Every fact below comes from Anthropic's and GitHub's own pages as I read them on 5 November 2026, and neither saw this text before publication.
What "Claude has already reviewed" can mean
Anthropic compares five ways to get a review in a table of its own, "Compare ways to get a review", and its commands page lists a sixth, /security-review, as a separate command. The first four columns below are in Anthropic's words; the last one comes from each path's own page.
| Option | What starts a review | Where it runs | Who can use it | What it leaves on the pull request |
|---|---|---|---|---|
| Code Review (managed) | "Pull request activity, as each repository is set, or an @claude review comment" |
"Anthropic infrastructure" | "Organizations on a Team or Enterprise plan with GitHub repositories, once an Owner turns it on" | "Findings arrive as inline comments on the lines concerned", "with a summary and a check run on the pull request"; the check run is named Claude Code Review |
/code-review |
"You run the command on your branch" | "Your Claude Code session" | "Anyone, on any plan" | findings in your session, since it "reports what it found in your conversation"; with --comment, it "posts the findings on a GitHub pull request as inline comments" |
| Ultrareview | "You start it from Claude Code" | "A cloud sandbox" | "Requires signing in with a claude.ai account."; its page adds "Ultrareview is a research preview feature." | for a pull request on github.com, "you can have Claude post the finished findings to the PR as a single plain comment from your own GitHub account"; for a branch, "When you review a branch, Claude Code shows the findings in your session only." |
| Claude in your own CI, with GitHub Actions or GitLab CI/CD | "Your pipeline's events, such as every pull request or merge request" | "Your own CI runners" | "Anyone with a Claude API key or cloud provider credentials, on GitHub or GitLab" | in the GitHub Actions review workflow with --comment, "Claude posts its review on the pull request, as an inline comment on each issue it finds or as one summary comment when it finds none."; without it, "Claude posts nothing, and you read the findings in the workflow run log." |
| Your own reviewer, as a subagent or skill | "You or Claude invoke it during a session" | "Wherever you run Claude Code" | "Anyone who writes one" | what you write it to do |
/security-review, a separate command on Anthropic's commands page |
not a row of Anthropic's comparison table | not a row of Anthropic's comparison table | not a row of Anthropic's comparison table | the commands page describes it as "Analyze the changes on your current branch for security vulnerabilities." |
A fresh context does not need a second tool. Anthropic writes that "In a terminal session, the review runs as a background subagent with its own context window", apart from the conversation you are in. The command has a second name, "/review is an alias of /code-review.", and Claude can start it when asked in plain words: "ask Claude to review your changes and it can run the skill". The GitHub Actions review workflow leaves some pull requests alone: "Claude skips draft and closed pull requests, pull requests it judges not to need a review, such as automated or trivial ones, and pull requests that already have a comment from Claude." Anthropic is direct about the limits of its managed review: "Code Review uses AI to review code and may miss bugs, security vulnerabilities, or other issues." and "A review with no findings doesn't mean your code is error-free or secure." Four of these paths are set against CodeRabbit, as of 17 October 2026, in an earlier comparison of Anthropic's review options with CodeRabbit.
What a Copilot review adds on the pull request
GitHub's default is a review on request: "By default, Copilot only reviews a pull request if you assign it to the pull request." The next sentence on the same page reads "However, you can configure automatic reviews.", either in the author's own settings or in a ruleset of the repository or organization. Without Review new pushes, GitHub writes, "Unless Copilot has been configured to review each push to a pull request, it will only review a pull request once." By default the review is a Comment review, which does not count toward required approvals.
What Copilot reads goes past the diff. Among its agentic capabilities GitHub lists "full project context gathering", and "This capability analyzes your entire repository to better understand the context of code changes." GitHub runs these capabilities on GitHub Actions and turns them on automatically for plans that include Copilot code review; if Actions is unavailable or its workflows fail, reviews still arrive but "will not include the additional features provided by the agentic capabilities." Two MCP servers come preset: "The GitHub MCP server and Playwright MCP server are enabled by default." Some files stay out of the review: "Dependency management files, such as package.json and Gemfile.lock", log files and SVG files.
For rules, Copilot reads .github/copilot-instructions.md and path-specific *.instructions.md files under .github/instructions/, and GitHub's advice is to "use AGENTS.md for standing rules you want to share across AI tools and agents". It takes them from the branch under review: "When reviewing a pull request, Copilot reads repository custom instructions, agent instructions, and agent skills from the head branch (the branch with your changes), not the base branch." Anthropic's managed Code Review takes its own rule file from the other side and reads REVIEW.md "from the pull request's base branch or from the default branch, never from the pull request's own changes."
Since 28 October 2026, "Copilot approvals, previously available in public preview, are now generally available."; "Approvals are off by default.", and once they are turned on, a Copilot approval satisfies the required-approval rule "the same way a teammate's approval would". With approvals on or off, "Every Copilot code review includes an approval assessment in the overview comment". GitHub's own advice on reading its reviewer is short: "Copilot is not guaranteed to spot all problems or issues in a pull request." and "Supplement Copilot's feedback with a human review."
Where the two reviews overlap
The overlap runs through rule files and through how much code each reviewer reads, and it differs by Claude path. GitHub's page on using code review says "Copilot code review also reads custom instructions from CLAUDE.md, GEMINI.md, and REVIEW.md files in your repository, if they exist." Anthropic's /code-review "follows your CLAUDE.md", and the same paragraph adds "It doesn't read REVIEW.md." The managed Code Review differs here: "The service reads both CLAUDE.md and REVIEW.md.", with REVIEW.md taken from the base or default branch, while Copilot takes its instructions from the head branch.
On code beyond the diff, each path is described in its own words. At Code Review, "several agents examine the diff and the surrounding code in parallel.", and a review "works from reading the diff and the surrounding code, without building or running your code." Claude, run as /code-review, "reviews the changes on your branch". Copilot "analyzes your entire repository" for the context of a change.
On models, Anthropic writes that "Code Review chooses the Claude models that reviews run on, and its admin settings don't offer a model choice." GitHub calls its reviewer "a purpose-built product that uses a carefully tuned mix of models, prompts, and system behaviors", and adds "Model switching is not supported". Model names for the Lite and Balanced levels were not found on GitHub's Copilot code review page, and provider names were not found in GitHub's changelog of 28 October 2026, as of 5 November 2026. That changelog describes Max as running "several independent review agents across models from different providers".
The shared ground is therefore partial: CLAUDE.md reaches Copilot and each of these two Claude paths, REVIEW.md reaches Copilot and, of these two, only the managed Code Review, which reads it from another branch, and how many findings overlap shows only when you count them.
When a second pass adds something, by the vendors' own pages
GitHub ties each effort level to a kind of change, and Anthropic names its own step for what a reading review does not do. Each row repeats a vendor's words; none is a verdict on either product.
| If this is your situation | What the pages say | What to try |
|---|---|---|
| Claude's review stayed in your terminal |
/code-review "reports what it found in your conversation"; Copilot posts its review on the pull request, a Comment review by default |
add --comment to /code-review, so its findings sit on the pull request next to Copilot's, then count |
| Claude's review is already on the pull request | Code Review: "Findings arrive as inline comments on the lines concerned"; Copilot reviews the same pull request on its own trigger | Copilot on request for the pull requests you choose, and the count below |
| The change is high-risk or release-critical | "Use Max for high-risk or release-critical pull requests that need the most thorough review." | Max for that pull request: "You can select Copilot review effort when requesting a review in the pull request" |
| The change is security-sensitive, multi-service, or in a repository with strict quality standards | "Use Balanced for security-sensitive code, multi-service pull requests, or repositories with strict quality standards." | Balanced, the default level; on the Claude side, /security-review, which "checks the diff for security vulnerabilities." |
| The change is routine | "Use Lite for routine changes where fast feedback is more important than exhaustive analysis." | Lite as the repository's default level, or no automatic Copilot review on such changes |
| You want reviewers built on more than one model provider | on GitHub's pages read for this text, models from different providers are described only for Max, which is "currently in public preview and subject to change." | Max where the change calls for it, with its preview status in mind |
| You want the code run, not only read | Code Review works "without building or running your code"; Anthropic describes /verify as "Confirm a code change does what it should by building your project's app, running it, and observing the result, rather than relying on tests or type checks." |
/verify on the change, next to whichever reviewers you keep |
For Max, GitHub describes several models working together: "Each model has its own strengths and blind spots, so combining them helps catch issues a single model might miss.", and "An aggregation agent then reconciles their findings, resolves disagreements, and double-checks accuracy before combining everything into a single review." Balanced, the level a review gets when nothing else is set, "Routes pull requests to a higher-reasoning model for longer analysis of complex logic, security-sensitive code, and cross-service changes (default)." For a change that has to be run and not only read, the step on Anthropic's commands page is /verify, whichever reviewers stay in the pipeline.
What the second pass costs
GitHub splits a Copilot review into "two cost components: AI credits for the model interaction (the review itself), and GitHub Actions minutes for the agentic capabilities (context gathering and tool use)." Its estimate for the credits is "On average, a review costs an estimated $0.25 USD with Lite, $1.50 USD with Balanced, and $10 USD with Max.", and the page bounds it at once: "These figures are estimates only.", "Actual costs can fall outside these ranges and may change as models evolve.", and "Estimates exclude GitHub Actions minutes." The typical range for Max is "$1.50 USD to $25 USD", and "Consumption generally increases with pull request size and repository custom instructions."
Who is billed depends on how the review started. "By default, when a repository automatically requests a code review, AI credits consumption is attributed to the pull request author." "If another user manually requests a review, consumption is attributed to that user instead." For organization members with paid Copilot licenses, GitHub writes that "the member's entitlement is billed by default", and an owner can choose to bill the organization instead.
On Anthropic's side, /code-review "counts toward your plan's usage or API billing the same way the rest of the session does." The managed Code Review "has no flat per-review price.", and "Reviews are billed to your organization, not to the developer who opened the PR or requested the review." Set to After every push, it "runs on each push, multiplying cost by the number of pushes".
A test on your own pull requests
The question behind this article describes one developer's pipeline: Claude Code's /code-review, then a second review in a fresh context, then a wait for Copilot on the pull request, and the question is whether that last pass earns its cost. The pages above say what each pass reads and how it is billed; only your own pull requests show what the last pass adds.
Take the last 10 pull requests that Claude and Copilot each reviewed, with Claude's findings written down or posted with --comment. For every finding, record three marks: which reviewer raised it, whether the other reviewer raised it too, and whether the team applied it. Then, for each reviewer, count the findings that only it raised and that were applied, and set that number next to what the reviewer cost over the same 10 pull requests: for Copilot, GitHub's average for your effort level or your own billing data; for Claude, your plan's usage or your organization's Code Review bill.
The count leaves three ordinary moves. Keep the pipeline as it is; turn Copilot's automatic review into a request for the pull requests that carry risk, as How to disable Copilot code review in GitHub? shows; or change Copilot's effort level for the repository. How many findings make a pass worth keeping is a line for your team to draw.
Sources and method
Every page below was read on 5 November 2026, each quotation was matched word for word against its page that day, and this text describes the pages as of that date.
GitHub: Copilot code review, using Copilot code review, changelog of 28 October 2026
Anthropic: Code Review, customizing Code Review, Code Review costs, /code-review, Ultrareview, GitHub Actions, commands
Top comments (0)