DeepSeek V4 Flash
DeepSeekRank #8 of 13 · 30 PRs · 35/95 golden bugs found
39.3
F1 · +1.6 vs avg
Recall
36.8%+1.3 vs avg
Precision
42.1%-1.9 vs avg
Cost / PR
$0.100
Cost / bug found
$0.08
Confidence
95% bootstrap interval on recall (2000 resamples over the 30 PRs):26.4–47.9points. This measures sampling variance from which PRs are in the set.
Judge noise (same submission, 3 independent re-scores): recall36.8 ± 0.0pts (n=3). This is the judge alone — the exact same findings, scored again.
Neither measures model run-to-run variance — re-running the review agent itself, not just the judge. One pass per entry.
Run configuration
HarnesskodusAccess pathapiExecution modereplayReasoningvendor-defaultJudgeclaude-haiku-4-5
Severity mix
What the model called its own findings — not recall by severity, goldens aren't severity-tagged.
Low 29Medium 49High 17Critical 0
Category mix
Only bug/performance/security are consistent across models — the rest is free text, bucketed as other.
Bug83Performance8Security4Other0