AI code review benchmark
How well does Mesrai catch real bugs in real pull requests? We're measuring it openly — with published scoring and every review linked.
Results coming soon
The Mesrai benchmark is in progress. Results will be published here, each verdict linked to the review it came from, so you can check every one yourself.
How we'll measure it
Real pull requests
Changes that look like everyday feature work, not toy snippets — each with an issue a careful reviewer should catch.
Bugs and false alarms
We measure both: whether a real problem is caught, and whether the reviewer stays quiet when the code is fine.
Published scoring
The scoring rules are written down before the run, so results can't be shaped afterwards.
Every review linked
Each verdict links to the actual review comments, so anyone can check the result for themselves.
Configuration disclosed
The model, rules and settings Mesrai ran with are published alongside the results.
Repeatable
We'll re-run the benchmark as Mesrai and the models it uses improve, and keep earlier results for comparison.
Benchmark — frequently asked
What will the Mesrai benchmark measure?
+How well Mesrai finds real problems in pull requests — bugs, security issues and performance problems — and how often it raises false alarms on code that is fine.
When will results be published?
+Soon. Results will appear on this page as soon as the run is complete and every verdict has been checked.
How will results be scored?
+With scoring rules published before the run. Each pull request gets a verdict, and every verdict links to the review it's based on.
Can I test Mesrai on my own code in the meantime?
+Yes. Every feature is free for 14 days with no credit card. Running Mesrai on your own pull requests is the most relevant test for your team.
Don't wait for the numbers — test it yourself.
14-day free trial with every feature. Run Mesrai on your own pull requests.
Start free