Qodo AI Code Review Benchmark (125 PRs)
Code
Qodo's AI Code Review Benchmark plants verified functional bugs and repository-specific best-practice violations in real, merged pull requests from production open-source repositories and has a model review each full pull request. This row covers Qodo's 125-pull-request results table, with one row per model and reasoning setting. Findings are matched against the planted issues, and the score is the F1 of recall over planted issues and precision over all reported findings.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5.5 | 56.8 |
| Claude Sonnet 5.5 | 54.0 |
| Claude Fable 5.1 | 53.5 |
| GLM-5.3 | 52.1 |
| DeepSeek V4 Pro 0813 | 49.9 |
| Kimi K3 | 49.0 |
| GPT-6 Sol | 48.9 |
| Gemini 3.8 Flash | 47.0 |
| GPT-6 Astra | 46.6 |
| GPT-6 Luna | 41.7 |
| Haiku 4.5 | 30.6 |