Rankings

AI Code Review Benchmark Leaderboard

Name: CodeReviewBench — AI Code Review Benchmark
License: https://opensource.org/licenses/MIT

30 real pull requests, 95 human-authored golden bugs, judged by claude-haiku-4-5. Ranked by F1 — recall alone rewards whoever talks most.

#	Model	F1	Precision	Recall	95% CI	$/PR	$/bug
01	DeepSeek V4 ProDeepSeek	43.9	43.6%	44.2%	32.3–56.3	$0.300	$0.21
02	Kimi K2.7 CodeMoonshot	43.1	50.0%	37.9%	28.1–48.7	$0.550	$0.46
03	Qwen3.8 MaxAlibaba	42.4	43.9%	41.0%	29.6–53.8	$1.210	$0.93
04	GLM-5.2Zhipu	40.2	47.0%	35.2%	24.7–47.1	$0.880	$0.79
05	GLM-5.3 FlashZhipu	40.0	41.0%	39.0%	29.1–49.4	$0.080	$0.07
06	Kimi K3Moonshot	39.9	38.7%	41.0%	32.3–50.5	$1.620	$1.25
07	DeepSeek V4 FlashDeepSeek	39.3	42.1%	36.8%	26.4–47.9	$0.100	$0.08
08	Muse Spark 1.2Meta	37.8	35.0%	41.0%	31.3–52.0	$0.490	$0.38
09	Qwen3.8 27BAlibaba	36.2	38.1%	34.4%	24.1–45.4	$0.340	$0.31
10	MiniMax M3MiniMax	29.6	35.6%	25.3%	16.5–36.4	$0.170	$0.21
11	Gemini 3.7 FlashGoogle	20.0	73.9%	11.6%	6.0–18.3	$0.270	$0.75

Next up

Not measured yet

The models people ask for most.