Learning center

Greptile vs CodeRabbit: settings, pricing and two matched reviews

Compare Greptile and CodeRabbit by when reviews run, the settings that shape them and how each one bills. Then read both tools' comments on two identical public pull requests, with the commit, permalinks and what the diff shows.

Proofline publishes this guide and sells a competing pull request reviewer. The vendor documentation and pricing linked below were checked on 20 November 2026. The two matched reviews come from public pull requests opened in July 2025, and both products have changed since.

How the two workflows differ#

Both tools review GitHub pull requests with repository context, take instructions from a file in the repository and offer a hosted trial. They differ most in when a review runs on its own and in what you pay for.

QuestionGreptileCodeRabbit
When does a review start on its own?When the pull request opens. Reviews on later pushes need autoReview set to include push (configuration reference)When the pull request opens, and again on each push, because incremental review is on by default. Automatic reviews pause after five reviewed commits (configuration reference)
Draft pull requestsSkipped unless triggerOnDrafts is set, or a comment asks for a draft review (developer guide)Skipped unless reviews.auto_review.drafts is set
Starting a review by handComment @greptileai, optionally with a question or a review tier@coderabbitai review reviews commits pushed since the last review; @coderabbitai full review reviews every file again (review commands)
How deep a review goesBase, Plus and Apex tiers, or Auto. Tier rules can pick a tier by target branch, label, path or files changed (review tiers)No review tiers. A profile sets how much feedback a review posts: quiet, chill (the default) or assertive
Narrowing the feedbackA severity threshold of 1 to 3, comment types (logic, syntax, style), instructions, ignored paths and label filtersPath-specific instructions, title keywords, label filters, base branches and ignored authors
Where settings livegreptile.json or .greptile/config.json, read from the pull request's source branch, or the dashboard.coderabbit.yaml in the repository, or the web settings
Running code during reviewTREX runs relevant code in a sandbox for 3 extra credits a run (TREX)CodeRabbit Verify starts your app in a prepared environment and tests the pull request (Verify)
Git hostsGitHub and several other hosts; enterprise server editions need Enterprise (code providers)GitHub and several other hosts, including self-hosted editions (GitHub Enterprise Server)

What you pay for#

Pricing termGreptileCodeRabbit
Billing unitPrepaid credits at $1 each, shared across the organization, with no seat feesPer developer: Essentials $24, Team $48 and Advanced $72 a month billed annually, or $30, $60 and $90 month to month
One reviewBase costs 1 credit, Plus 3 and Apex 10. With Auto you pay for the tier that ran. Only completed reviews spend credits (billing)Each plan has an hourly review allowance per developer: 5, 8 and 10 reviews. Incremental reviews after a push and manual commands each use one
Beyond the allowanceNew reviews pause when credits run out, unless automatic recharge is onWith usage-based reviews turned on, continued reviews cost $0.25 per reviewed file (pricing)
Free useStarter: one active developer and 50 credits a month (pricing)Free plan summarizes pull requests; reviews run in the editor extension and CLI
Trial14-day Pro trial14-day trial with Advanced features
Open-source projectsFree for qualified non-commercial projects under an OSI-approved license, by applicationPublic repositories are reviewed free, with rate limits that vary by repository
Self-hostingEnterpriseEnterprise

Neither unit converts directly into the other. A seat covers one developer's reviews up to an hourly rate; a credit covers one review at one depth. Count your pull requests, pushes per pull request and the share of large or release changes during a trial, then price both plans from those counts.

When to choose each tool#

Greptile: set review depth for each pull request#

Consider a team that ships routine changes all week and a release pull request into main every Friday, with contributors who open only a few pull requests a month. Greptile's tier rules let that team run Base reviews on routine work and Apex on anything targeting main or touching more than 30 files. The credits follow the reviews, so an occasional contributor costs nothing until a review runs.

Check how your team pushes fixes. By default Greptile reviews when the pull request opens, so either add push to autoReview or ask for a review in a comment after a substantial change. Each of those reviews spends credits.

CodeRabbit: review every push under a seat price#

Consider a team of steady contributors who push several rounds of fixes to each pull request and want each round checked before a human looks again. CodeRabbit reviews the new commits on each push by default, and path instructions let the team give migrations, API handlers and tests their own guidance. The monthly charge follows the number of developers.

Check how the incremental model fits your comparison. A review command covers only commits since the last review, and automatic reviews pause after five reviewed commits until someone resumes them. When you compare findings across a whole pull request, use full review or collect the findings that still apply from every pass.

Two matched reviews of the same commit#

Matched public examples are rare: few public repositories run both tools, and a maintainer seldom records which finding led to a fix. The usable pairs we found come from Greptile's own benchmark, run in July 2025. Greptile copied five open-source repositories into one GitHub organization, with a separate copy of each repository for each tool, and opened the same bug-introducing change in every copy. It says each tool ran on its hosted plan with default settings and no custom rules, and that it counted a bug only when a line-level comment pointed to the faulty code and explained the impact.

We checked that both copies of each pull request point to the same head commit and base branch, read the diff, and followed every comment linked below. Nobody from the original projects reviewed these pull requests. They remain open and unmerged, so they show no maintainer decision. Greptile scored them itself, and it sells one of the two tools.

ChangeHead commitGreptile's line commentCodeRabbit's line commentBenchmark score
Keycloak: group permissions for fine-grained admin679f44692dFlags canManage() for accepting the VIEW scopeFlags the same check and suggests removing VIEWBoth found it
Cal.com: insights queries rewritten with Prisma.sql56ff75f07bSays combining conditions may produce invalid SQLNone about injectionGreptile found it; CodeRabbit missed it

Keycloak: a manage check that accepts view permission#

Both copies of pull request 6 hold head commit 679f44692dbc288bedc8f1535071c8467dbaf70c on base branch feature-groups-authz-baseline. The change adds GroupPermissionsV2. Its canManage() method passes VIEW and MANAGE to a helper that returns true when the caller holds either scope, so an administrator with only view permission on all groups would pass the manage check. The corresponding method on Keycloak's current main branch checks only MANAGE.

Both tools posted a line comment on that method. Greptile's review, dated 28 July 2025, said: "Inconsistent scope check: canManage() checks VIEW and MANAGE scopes, but should only check MANAGE scope." CodeRabbit said: "The method is checking for VIEW scope when it should only check for MANAGE scope." It attached a suggested change that drops VIEW.

The timing differs. CodeRabbit's walkthrough is dated 26 July. Its review with the line comment is dated 30 July, after a @coderabbitai review comment that day. CodeRabbit's output records the settings it used: the chill profile, on the plan then called Pro and now called Essentials.

Cal.com: an injection label the diff does not support#

Both copies of pull request 5 hold head commit 56ff75f07b92ba8ef499ff28a2f98c25f339fdae on base branch insights-query-foundation. The change replaces Prisma's typed filters with SQL fragments built with Prisma.sql. The benchmark labels it a critical SQL injection risk. In the diff, every value enters through a Prisma.sql template placeholder, which Prisma sends as a query parameter (Prisma's guidance on injection). The diff contains no unsafe raw query method and no string concatenation into SQL.

Greptile's review summary says the conversion "introduces potential SQL injection vulnerabilities" but points to no line. Its one line comment says the composition "may result in invalid SQL like (1=0) AND (other_condition)". That expression is valid SQL and returns no rows, which is what an unauthorized caller should get. The benchmark does not count summary-only mentions, so we could not reproduce its score for Greptile from the evidence on the page.

CodeRabbit's review posted no actionable comment. Among its additional comments, an item titled "Potential authorization bypass when auth returns NOTHING_CONDITION" carries CodeRabbit's own note, "Likely an incorrect or invalid review comment". Another item says the filter code "Safely parameterizes values using Prisma.sql". Neither tool identified a defect we could confirm in this change.

What the two pairs establish#

On the Keycloak change, both tools identified the faulty check in a line comment. On the Cal.com change, neither line comment identified a defect we could confirm. Two pull requests from a vendor-run set cannot rank the tools, and the results describe the July 2025 products. A comparison of defects found, cost per review or review time needs your own pull requests, run on the same commits with recorded settings.

To check that both copies of a pull request hold the same commit, list their heads with Git:

git ls-remote https://github.com/ai-code-review-evaluation/keycloak-greptile refs/pull/6/head
git ls-remote https://github.com/ai-code-review-evaluation/keycloak-coderabbit refs/pull/6/head

Run your own matched comparison#

  1. Choose pull requests you may share with both vendors: a known bug, a change that reaches another module, and an ordinary change that should draw little feedback.
  2. Open each change against the same base and head commit for both tools, with the settings you would use in production. Write the settings down: Greptile's tier, severity threshold and comment types; CodeRabbit's profile and path instructions.
  3. Let each tool run its automatic review, then record any manual commands you add. For CodeRabbit, compare a full review or every pass through the same commit.
  4. Save the permalink of each comment you judge. Decide whether each finding is a confirmed defect, a preference or a claim the evidence does not support, before you read the other tool's output.
  5. Record what the trial charged: Greptile credits by tier, or CodeRabbit reviews against the hourly allowance and any usage-based charges.

The AI code review tools guide has a scorecard for this trial and covers GitHub Copilot and Qodo as well.

Where Proofline fits#

Proofline reviews GitHub pull requests on the exact commit and cites the code behind each finding. To compare a third reviewer on the same changes, compare CodeRabbit and Proofline or Greptile and Proofline. New organizations can try one calendar month of hosted code review on private and public repositories, subject to eligibility and fair use.

Questions#

Is Greptile or CodeRabbit better?

Neither vendor's published numbers settle it, and each tool fits a different workflow. Greptile lets you set review depth for each pull request and pay per review. CodeRabbit reviews each push by default and charges per developer. Run both on the same commits from your own repositories and judge the findings yourself.

Does each tool review again after a push?

CodeRabbit does by default, reviewing only the new commits. Greptile reviews when the pull request opens unless you add push to autoReview or ask for a review in a comment.

Which one costs less?

It depends on your team's pull request volume, pushes per pull request and review tiers. Greptile charges per completed review at the tier that ran. CodeRabbit charges per developer and adds per-file charges only after the hourly allowance, when usage-based reviews are on. Price both from the counts you record during a trial.

One month of free code review.

New organizations get one calendar month of hosted code review for private and public GitHub repositories. No card or cloud connection required. Eligibility and fair-use terms.

Start free with GitHub

Related reading