Benchmarks · safety

DeepsecBench-AA

DeepsecBench-AA (based on Vercel's DeepsecBench), part of the AA Cyber Index: finding and fixing vulnerabilities in real codebases; score.

19 models · 19 results · data as of 2026-10-29 · artificialanalysis.ai ↗

Breakdown

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

Model
GPT-6 Sol max	48,727,649	94,055	367,989	46,517,146
GPT-6 Astra max	14,977,455	86,072	178,999	13,181,687
DeepSeek V4.1 Flash max	103,034,568	149,047	1,045,635	100,642,557
Grok 4.7 xhigh	75,864,787	123,622	687,093	73,895,643
Opus 5.5 max, fallback	77,506,876	165,093	820,735	75,417,130
MiMo V2.6 Pro	18,370,759	81,514	408,145	17,446,639
Opus 5 max	48,358,507	141,258	524,357	46,686,711
GPT-5.6 Terra max	61,921,170	140,399	440,031	58,026,935
DeepSeek V4.1 Flash 1017 max	77,691,277	133,268	755,443	75,766,994
GLM 5.3 max	48,824,226	117,905	547,778	47,502,397
GPT-6 Luna max	30,861,681	64,008	659,638	28,996,658
Fable 5.1 max, fallback	51,927,709	133,798	456,456	50,382,927
GLM 5.3 Flash max	20,748,986	76,530	321,327	19,877,027
Gemini 3.8 Flash high	46,579,706	39,530	521,664	45,319,851
Muse Spark 1.3 xhigh	16,970,292	60,809	165,344	16,206,702
Kimi K3 max	3,554,553	39,516	102,737	3,202,851
Qwen 3.8 27B xhigh	20,991,269	78,973	322,538	20,133,499
Qwen 3.8 2.4T-A95B	17,541,247	61,641	314,023	16,699,589
MiniMax M3	82,856,518	165,638	317,760	81,801,126

All results

19 rows

#	Model	Effort	DeepsecBench-AA
1	GPT-6 Sol	max	46.3%
2	GPT-6 Astra	max	36.9%
3	DeepSeek V4.1 Flash	max	31.2%
4	Grok 4.7	xhigh	26.9%
5	Opus 5.5	max	26.9%
6	MiMo V2.6 Pro		26.2%
7	Opus 5	max	25.9%
8	GPT-5.6 Terra	max	25.1%
9	DeepSeek V4.1 Flash 1017	max	24.9%
10	GLM 5.3	max	24.3%
11	GPT-6 Luna	max	23.5%
12	Fable 5.1	max	23.5%
13	GLM 5.3 Flash	max	19.9%
14	Gemini 3.8 Flash	high	14.5%
15	Muse Spark 1.3	xhigh	12.5%
16	Kimi K3	max	12.5%
17	Qwen 3.8 27B	xhigh	11.5%
18	Qwen 3.8 2.4T-A95B		9.5%
19	MiniMax M3		1.1%