Open Code Review AACR-Bench: leaderboard
Alibaba's Open Code Review runs AACR-Bench — 200 real pull requests from 50 open-source repositories in ten languages, with 1,505 expert-verified review issues — through three review harnesses (Open Code Review, Claude Code, Codex) and scores each model-plus-harness pair by F1 between the comments it posts and the annotated issues. The site's headline column; higher is better.
Metric: F1 (%). Source: open-codereview.ai. Saturation forecast: Not forecast. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-5.2 | 21.3 | #80 |
| 2 | GPT-5.5 | 21 | #33 |
| 3 | GLM-5.1 | 20.4 | #109 |
| 4 | DeepSeek V4 Pro | 17.9 | #112 |
| 5 | Claude Opus 4.8 | 14.13 | #39 |
| 6 | Qwen 3.7 Max | 12.17 | #85 |
| 7 | Claude Opus 4.6 | 11.57 | #67 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=open-code-review-aacr-bench · How It Works · Data refreshed daily, snapshot 2026-11-05.