Review every merge request. For 25 cents.
#5 of 20 on the public Martian Code Review Benchmark, at a flat $0.25 per review. No token math, no bill surprises, no rationing reviews to fit a budget.
#What you'll actually see in your MRs.
We measure GitLab Duo Code Review on real merge requests against human-annotated ground truth. Here's what that means in practice.
Catches about half of real issues
On the independently-scored Martian Code Review Benchmark, GitLab Duo Code Review surfaces roughly half the issues a human reviewer would on real MRs. Some MRs get full coverage. Others get a partial pass. Use it as a fast first pass on top of human review, not as a replacement.
Designed to run on every MR
Many AI review tools use variable, token-based pricing. Bigger MRs cost more; monthly bills swing with what your team happens to ship. GitLab Duo Code Review uses flat-rate pricing instead: $0.25 per MR, same price every time. The predictability lets teams review everything, not just the changes that fit the budget.
Tested on the languages you ship in
GitLab Duo Code Review was measured on real merge requests across 10 languages: Python, JavaScript, TypeScript, Go, Java, C, C#, Swift, Rust, and Ruby. The codebases span web frameworks, systems code, mobile apps, developer tools, and full-stack platforms. If you ship in any of these, the results apply to you.
Steered by your conventions
A YAML file in your repo tells GitLab Duo Code Review what to focus on. Security, performance, maintainability, or your team's coding standards. Different rules per file pattern. The review adapts to your repo, not the other way around.
#Designed for the way you ship.
Six things that make GitLab Duo Code Review feel like a teammate, not a tool to configure.
One @mention to start
Assign @GitLabDuo as a reviewer, or use /request_review. The review begins immediately on the MR you're already in.
Review every MR automatically
Turn on auto-review at the project, group, or instance level. Cascading defaults: set it once at the group, every new project inherits.
Reads your codebase, not just the diff
A pre-scan finds the related files and context first. The full review runs with your directory structure, tests, and cross-file dependencies in scope. Feedback understands your code, not just the lines that changed.
Steer it with custom instructions
A mr-review-instructions.yamlfile tells the reviewer what to focus on per file pattern: security, performance, or your team's standards.
Interactive, not one-shot
Reply to a comment, @mention the reviewer back, ask for alternatives. The conversation continues threaded in the MR, like a teammate would.
Free tier · flat-rate pricing
Available on the GitLab Free tier through GitLab Duo Agent Platform. Flat $0.25 per review, every MR, every project, every time. No tokens to count.
See the full feature reference
Activation, configuration, model selection, supported flows, and limits.
We don't grade our own homework.
The Martian Code Review Benchmark scores AI code review tools on real MRs against expert-verified golden comments. Martian runs the evaluation and publishes the leaderboard. GitLab Duo Code Review is listed at #5 of 20.
Two numbers matter for AI code review: precision (how often a comment is correct) and recall (how many real issues are caught). F1 combines them. Higher is better.
| Rank | Tool | F1 |
|---|---|---|
| #1 | Cubic Dev | 61.8% |
| #2 | Qodo Extended | 57.9% |
| #3 | Augment | 53.5% |
| #4 | Qodo | 48.4% |
| #5 | GitLab Duo Code Reviewus | 48.3% |
| #6 | macroscope | 46% |
| #7 | Cursor Bugbot | 45.5% |
| #8 | Devin | 44.2% |
| #9 | Gemini | 44% |
| #10 | Greptile | 44% |
Methodology · model comparison · key findings.
Review every MR for $0.25.
Flat-rate pricing on the GitLab Free tier. To see it on your own codebase, assign @GitLabDuo to your next MR or use /request_review. The review begins immediately.