DIY GPT-5.5 code review team beats Cursor Bugbot on a 50-PR benchmark
s0ck_r4w · hn · 2026-07-24
What the post shows
A DIY code review team built from three GPT-5.5 reviewers — bug-hunter, keeper, and sweeper — beat Cursor Bugbot and CodeRabbit on Martian’s Open Code Review Bench.
Key details
- The setup is intentionally simple: the three reviewers are all the same model, and the only difference is their review focus.
- The author then runs the team three times and keeps only findings that appear in all three runs.
- On 50 real PRs, the system reached 0.463 F1, ahead of Cursor Bugbot (0.445) and CodeRabbit (0.352).
- The main takeaway is that customizing a review workflow to a codebase may matter more than elaborate personas.
Why it matters
The post argues that building your own review stack on top of existing harnesses is still underexplored, and that off-the-shelf review tools may not be enough for teams that want tighter codebase-specific tuning.
More from coding & agent
- Stanford's PaperFormalizer turns research papers into Lean-verified versions with author in the loop — StanfordUniversity · 2026-10-27
- Hugging Face TRL makes score centering + masked importance sampling default in async GRPO — _lewtun · 2026-10-27
- AutoResearch agent's weird training curve gets dug into instead of discarded — ZimingLiu11 · 2026-10-27
- Firecrawl offers 2.5x credits to a16z Speedrun founders, unveils $75M Series B — devdigest · 2026-10-27
- Chums: open-source Apple Watch app puts Claude Code and Codex on your wrist — brandon_galang · 2026-10-27
- Open-source Clairic Studio turns pasted text into typed data and working code in 5 languages — dfinke · 2026-10-27