Skip to content

AI model leaderboards

Get API keyRead the docs
See top models, labs, apps, and providers on AI Gateway with token, request, and cost trends.

DeepsecBench measures how well different AI models find security vulnerabilities in application code, based on the deepsec cyber harness(opens in new tab). The report combines recall and precision into a single benchmark score.

Score vs Cost

010203040$0$50$100$150Cost (USD)Score

Score vs Total Time

0102030400h1h2h3h4h5hTotal timeScore

Results

1
GPT-6 Astra xhighopenai/gpt-6-astra
37.79
32.8%97.9%2$63.7049m 47s1,683.9Codex
2
GPT-5.6 Sol xhighopenai/gpt-5.6-sol
35.44
30.6%96.3%3$55.983h 39m 02s2,274.9Codex
3
Claude Opus 5 maxanthropic/claude-opus-5
32.44
28.0%88.0%10$127.932h 34m 19s4,453.7Claude Code
4
GPT-6 Astra mediumopenai/gpt-6-astra
29.33
25.0%95.5%3$35.2021m 49s973.0Codex
5
Claude Opus 5 mediumanthropic/claude-opus-5
28.25
24.6%70.5%26$31.9647m 01s999.7Claude Code
6
GPT-5.6 Luna xhighopenai/gpt-5.6-luna
26.68
22.8%81.2%13$24.593h 01m 40s3,343.4Codex
7
GPT-5.6 Sol mediumopenai/gpt-5.6-sol
25.00
21.1%94.3%3$17.9530m 52s703.0Codex
8
GLM-5.3 highzai/glm-5.3
21.91
18.5%80.7%11$16.222h 05m 37s1,232.1Pi
9
GPT-5.5 xhighopenai/gpt-5.5
21.11
17.7%95.5%2$43.422h 22m 57s2,158.8Codex
10
GLM-5.3 maxzai/glm-5.3
20.39
17.2%75.9%13$14.601h 29m 34s1,241.8Pi
11
GPT-5.6 Terra xhighopenai/gpt-5.6-terra
19.14
15.9%95.3%2$27.811h 45m 32s2,108.7Codex
12
Kimi K3 xhighmoonshotai/kimi-k3
17.49
14.7%77.1%11$12.381h 59m 39s731.9Pi
13
Claude Sonnet 5 maxanthropic/claude-sonnet-5
16.99
14.2%76.1%11$45.173h 08m 20s4,601.8Claude Code
14
Deepseek V4 Flash xhighdeepseek/deepseek-v4-flash
16.47
13.8%74.0%13$5.941h 18m 09s1,481.0Pi
15
Qwen3.8 Max xhighalibaba/qwen3.8-max
16.47
13.8%73.9%12$6.331h 52m 20s342.6Pi
16
Grok 4.5 mediumxai/grok-4.5
16.47
13.8%73.3%12$11.041h 37m 33s909.6Pi
17
Grok 4.6 xhighxai/grok-4.6
16.16
13.4%100.0%0$22.294h 02m 36s3,456.4Pi
18
Qwen3.8 Max mediumalibaba/qwen3.8-max
15.53
12.9%78.9%8$6.222h 03m 33s383.4Pi
19
Grok 4.5 xhighxai/grok-4.5
15.52
12.9%77.8%10$5.601h 24m 17s1,010.5Pi
20
Deepseek V4 Flash mediumdeepseek/deepseek-v4-flash
15.42
12.9%66.7%16$5.061h 21m 16s1,422.5Pi
21
GPT-5.5 mediumopenai/gpt-5.5
12.58
10.3%92.3%2$14.9424m 05s751.4Codex
22
GPT-5.6 Terra mediumopenai/gpt-5.6-terra
12.07
9.9%92.0%2$5.1513m 15s617.7Codex
23
GPT-5.6 Luna mediumopenai/gpt-5.6-luna
12.03
9.9%82.8%5$5.2311m 59s572.0Codex
24
Gemini 3.6 Flash mediumgoogle/gemini-3.6-flash
11.51
9.5%80.0%6$10.7426m 21s2,730.7Pi
25
Kimi K3 mediummoonshotai/kimi-k3
11.43
9.5%64.1%14$10.841h 29m 16s676.6Pi
26
GLM-5.2 mediumzai/glm-5.2
10.42
8.6%62.5%12$11.091h 03m 38s1,498.9Pi
27
GLM-5.2 xhighzai/glm-5.2
9.86
8.2%53.8%18$51.842h 30m 44s4,727.6Pi
28
Gemini 3.6 Flash xhighgoogle/gemini-3.6-flash
8.95
7.3%77.3%5$5.3322m 18s1,822.4Pi
29
Claude Opus 4.8 maxanthropic/claude-opus-4.8
6.89
5.6%81.3%3$7.5619m 29s80.8Claude Code
30
Claude Opus 4.8 mediumanthropic/claude-opus-4.8
6.85
5.6%61.9%8$10.8118m 16s78.1Claude Code
31
Inkling xhighthinkingmachines/inkling
6.30
5.2%48.0%13$1.9311m 34s300.9Pi
32
Claude Sonnet 5 mediumanthropic/claude-sonnet-5
5.81
4.7%61.1%7$6.6615m 47s442.2Claude Code
33
Inkling mediumthinkingmachines/inkling
5.24
4.3%37.0%17$2.1810m 31s509.0Pi
34
Claude Haiku 4.5 defaultanthropic/claude-haiku-4.5
4.71
3.9%33.3%18$5.9129m 40s912.4Claude Code
35
Gemini 3.5 Flash Lite mediumgoogle/gemini-3.5-flash-lite
2.13
1.7%44.4%5$0.5801m 46s459.9Pi
36
Gemini 3.5 Flash Lite xhighgoogle/gemini-3.5-flash-lite
1.07
0.9%28.6%5$0.1801m 07s406.5Pi