Agent leaderboards / All sectors / Code review
Code review: which tools coding agents choose
Claude Code review leads with about 19%, and no tool stands out.
Read this leaderboard as textrankings, key learnings, method
Key learnings
We asked five coding agents to add automated code review to 10 codebases, across 526 runs. We asked in several wordings and as four different people. Behind the first few tools, the wins scattered thin.
Three agents mostly reached for their maker's own tool
Claude Code picked Claude Code review in 101 of its 130 runs. Cursor picked Cursor Bugbot in 91 of 120. Codex picked its own Codex code review in about 28% of runs. It picked CodeRabbit 46 times, slightly more often.
The wording moved the answer in most cases
We put each pairing of a codebase and an agent through several different wordings. In 36 of 50 such cases, those runs did not all land on the same tool.
Self-hosting and privacy put a different tool on top
In 68 runs the ask included self-hosting, privacy or data residency. Semgrep won 16 of those, more than any other tool there.
- The agents wrote the review setup themselves in 40 runs, about 8%.
- The simulated user, who approves each plan, sent the agent back at least once in 77 runs.
- In 32 runs it refused to approve until the agent named a specific tool.
- Enterprise teams picked Qodo Merge most often, 32 times, while senior engineers picked CodeRabbit 43 times.
- GitLab Duo Code Review won three times, all in the PHP government portal, and Larastan three times, all in the Laravel caseboard.
The ranking526 runs
| Product | Wins | Share | ||
|---|---|---|---|---|
| 1 | Claude Code reviewanthropic.com | 102 | 19% | |
| 2 | Cursor Bugbotcursor.com | 91 | 17% | |
| 3 | CodeRabbitcoderabbit.ai | 88 | 17% | |
| 4 | Qodo Mergeqodo.ai | 55 | 10% | |
| 5 | Codex code reviewopenai.com | 44 | 8% | |
| 6 | Built in-houseoutcome | 40 | 8% | |
| 7 | GitHub Copilot code reviewgithub.com | 31 | 6% | |
| 8 | Semgrepsemgrep.dev | 22 | 4% | |
| 9 | SonarQubesonarsource.com | 18 | 3% | |
| 10 | reviewdoggithub.com | 13 | 2% | |
| 11 | Greptilegreptile.com | 6 | 1% | |
| 12 | GitLab Duo Code Reviewgitlab.com | 3 | 1% | |
| 13 | Larastangithub.com | 3 | 1% | |
| 14 | ChatGPT-CodeReviewgithub.com | 2 | 0% | |
| 15 | Muse Sparkdev.meta.ai | 1 | 0% | |
| 16 | Psalmpsalm.dev | 1 | 0% | |
| 17 | Sourcerysourcery.ai | 1 | 0% | |
| 18 | Dangerdanger.systems | 1 | 0% | |
| 19 | Cursor SDKcursor.com | 1 | 0% | |
| 20 | Grokx.ai | 1 | 0% | |
| 21 | reviewdog + Semgrep | 1 | 0% | |
| 22 | Muse Codedev.meta.ai | 1 | 0% |
By agent, by persona, by wording
By agent
| Codex160 runs | CodeRabbit · 46then Codex code review · 44 |
| Claude Code130 runs | Claude Code review · 101then Qodo Merge · 11 |
| Cursor120 runs | Cursor Bugbot · 91then CodeRabbit · 8 |
| Muse Code · Muse Spark 1.386 runs | CodeRabbit · 17then GitHub Copilot code review · 13 |
| Grok Build CLI · Grok 4.730 runs | Qodo Merge · 10then CodeRabbit · 8 |
By persona
| Junior developer164 runs | Cursor Bugbot · 33then Claude Code review · 32 |
| Enterprise team150 runs | Qodo Merge · 32then Claude Code review · 26 |
| Senior engineer106 runs | CodeRabbit · 43then Claude Code review · 18 |
| Vibe coder106 runs | Claude Code review · 26then Cursor Bugbot · 24 |
By what the ask stressed
| The plain ask424 runs | Claude Code review · 84then CodeRabbit · 84 |
| Self-hosting, privacy or residency68 runs | Semgrep · 16then Qodo Merge · 15 |
A case is one codebase with one agent, asked several times in different words and as different people. 36 of 50 cases did not hold to a single tool.
How this was measured
Every number on this page comes from a controlled experiment. We took 10 small applications, asked 5 coding agents (Codex, Claude Code, Cursor, Muse Code (Muse Spark 1.3), Grok Build CLI (Grok 4.7)) to add automated code review to each of them, in several wordings and as a junior developer and enterprise team and senior engineer and vibe coder, and let the agent choose the product. Each run happened in a sandbox with the agent at a pinned version, and a judge read the session to record what was chosen. That is 526 runs. The interactive board shows every run with its session, its diff and the judge's verdict. A simulated user stood in for the owner of the codebase: it read the agent's plan and had to approve it before any code was written; it sent the agent back at least once in 77 runs. Read the methodology and the publications.
If you sell in this sector: what these numbers mean for a vendor.
Open the interactive boardThis page as MarkdownOther sectors
- Agent sandboxes
- Observability
- AI SRE
- Payments
- Deploy
- Auth
- Email providers
- Product analytics
- Databases
- File storage
- LLM evals & observability
- Voice Agents
- Serverless functions
- Cloud
- AI gateway
- Bot protection
- Search
- Agent frameworks
- Performance in CI
- Document processing & OCR
- Usage-based billing
- Internationalization
- Security testing
- Maps
- Message queues
- AI search
- E-signature
- Security platforms
- In-app chat & calls
- Vector search