Benchmarks · safety

CWE-Bench

CWE-Bench: security-weakness detection; score.

17 models · 17 results · data as of 2026-10-29 · artificialanalysis.ai ↗

Breakdown

darker is better within each column · click a column to sort

Model
DeepSeek V4.1 Flash 1017 max	2,436,663,601	4,280,718	12,341,961	2,408,224,443
DeepSeek V4.1 Flash max	4,594,952,903	5,402,028	25,578,446	4,549,906,573
Grok 4.7 xhigh	1,078,951,728	2,904,314	10,298,053	1,054,267,901
GPT-6 Sol max	573,330,129	1,836,949	5,358,441	556,749,967
GPT-6 Astra max	199,940,815	1,665,387	3,218,600	187,067,854
MiMo V2.6 Pro	654,516,975	2,369,361	9,238,601	637,679,278
Muse Spark 1.3 xhigh	920,449,199	2,574,168	4,785,302	902,119,245
Opus 5.5 max, fallback	1,578,881,192	4,829,828	12,378,798	1,549,262,852
GPT-6 Luna max	698,411,140	1,542,437	6,491,740	682,367,478
GLM 5.3 Flash max	1,107,384,820	3,313,960	7,089,693	1,090,257,431
GLM 5.3 max	1,355,186,150	3,510,448	7,972,871	1,335,719,333
Kimi K3 max	321,383,607	1,695,960	2,881,955	312,963,385
Gemini 3.8 Flash high	2,195,266,049	2,852,603	13,863,084	2,164,844,080
Fable 5.1 max, fallback	955,120,873	4,287,928	7,620,352	933,755,516
MiniMax M3	1,509,816,528	3,642,932	3,607,627	1,494,606,573
Qwen 3.8 2.4T-A95B	786,616,863	2,699,409	6,235,797	771,482,547
Qwen 3.8 27B xhigh	907,507,966	3,131,298	6,871,992	891,250,949

All results

17 rows

#	Model	Effort	CWE-Bench
1	DeepSeek V4.1 Flash 1017	max	69.2%
2	DeepSeek V4.1 Flash	max	68.3%
3	Grok 4.7	xhigh	68.3%
4	GPT-6 Sol	max	64.2%
5	GPT-6 Astra	max	63.3%
6	MiMo V2.6 Pro		63.3%
7	Muse Spark 1.3	xhigh	60.8%
8	Opus 5.5	max	58.3%
9	GPT-6 Luna	max	56.7%
10	GLM 5.3 Flash	max	55.8%
11	GLM 5.3	max	55.8%
12	Kimi K3	max	53.3%
13	Gemini 3.8 Flash	high	52.5%
14	Fable 5.1	max	50.8%
15	MiniMax M3		41.7%
16	Qwen 3.8 2.4T-A95B		28.3%
17	Qwen 3.8 27B	xhigh	26.7%