# Artificial Analysis DeepsecBench-AA (DeepsecBench-AA)

> Artificial Analysis' implementation of Vercel's DeepsecBench, testing agents on finding security vulnerabilities in real web application code. One of the three components of the AA Cyber Index.

Canonical page: https://benchlm.ai/benchmarks/aadeepsecbench

- Category: [Agentic](/agentic)
- Last updated: October 29, 2026

## About DeepsecBench-AA

- Year: 2026
- Tasks: Vulnerability discovery in real codebases
- Format: Pass@1
- Difficulty: Security code review
- Paper: [DeepsecBench-AA Benchmark Leaderboard](https://artificialanalysis.ai/evaluations/deepsecbench-aa)

Independently run by Artificial Analysis under its published harness. Display-only; not yet admitted as independent-run evidence.

DeepsecBench-AA is currently displayed on BenchLM for reference, but it is excluded from the weighted scoring formula.

## Leaderboard (21 models)

| Rank | Model | Creator | Score |
|------|-------|---------|-------|
| 1 | [GPT-6 Sol](/models/gpt-6-sol) | OpenAI | 46.3% |
| 2 | [GPT-6 Astra](/models/gpt-6-astra) | OpenAI | 36.9% |
| 3 | [Kimi K3.1 Pro](/models/kimi-3-1-pro) | Moonshot AI | 36.7% |
| 4 | [DeepSeek V4.1 Flash 1017](/models/deepseek-v4-1-flash) | DeepSeek | 31.2% |
| 5 | [Claude Opus 5.5](/models/claude-opus-5-5) | Anthropic | 26.9% |
| 6 | [Grok 4.7](/models/grok-4-7) | xAI | 26.9% |
| 7 | [MiMo-V2.6-Pro](/models/mimo-v2-6-pro) | Xiaomi | 26.2% |
| 8 | [Claude Opus 5](/models/claude-opus-5) | Anthropic | 25.9% |
| 9 | [GPT-5.6 Terra](/models/gpt-5-6-terra) | OpenAI | 25.1% |
| 10 | [GLM-5.3](/models/glm-5-3) | Z.AI | 24.3% |
| 11 | [Claude Fable 5.1](/models/claude-fable-5-1) | Anthropic | 23.5% |
| 12 | [GPT-6 Luna](/models/gpt-6-luna) | OpenAI | 23.5% |
| 13 | [GLM-5.3-Flash](/models/glm-5-3-flash) | Z.AI | 19.9% |
| 14 | [Mistral Large 4](/models/mistral-large-4) | Mistral | 16.0% |
| 15 | [Step 5 Preview](/models/step-5-preview) | StepFun | 15.8% |
| 16 | [Gemini 3.8 Flash](/models/gemini-3-8-flash) | Google | 14.5% |
| 17 | [Kimi K3](/models/kimi-k3) | Moonshot AI | 12.5% |
| 18 | [Qwen3.8-27B](/models/qwen3-8-27b) | Alibaba | 11.5% |
| 19 | [Beam 1.0 501B-A23B](/models/reflection-beam) | Reflection AI | 6.9% |
| 20 | [MiniMax M3](/models/minimax-m3) | MiniMax | 1.1% |
| 21 | [Nemotron 3 Ultra](/models/nemotron-3-ultra) | NVIDIA | 1.0% |

## FAQ

### What does DeepsecBench-AA measure?

Artificial Analysis' implementation of Vercel's DeepsecBench, testing agents on finding security vulnerabilities in real web application code. One of the three components of the AA Cyber Index.

### Which model scores highest on DeepsecBench-AA?

GPT-6 Sol by OpenAI currently leads with a score of 46.3% on DeepsecBench-AA.

### How many models are evaluated on DeepsecBench-AA?

21 AI models have published results on DeepsecBench-AA in the BenchLM catalog.

### Does DeepsecBench-AA affect BenchLM's overall score?

Not directly. DeepsecBench-AA is still displayed on BenchLM for reference, but it is excluded from the weighted scoring formula.

## Compare Top Models on DeepsecBench-AA

- [GPT-6 Sol vs GPT-6 Astra](/compare/gpt-6-astra-vs-gpt-6-sol)
- [GPT-6 Astra vs Kimi K3.1 Pro](/compare/gpt-6-astra-vs-kimi-3-1-pro)
- [Kimi K3.1 Pro vs DeepSeek V4.1 Flash 1017](/compare/deepseek-v4-1-flash-vs-kimi-3-1-pro)
- [DeepSeek V4.1 Flash 1017 vs Claude Opus 5.5](/compare/claude-opus-5-5-vs-deepseek-v4-1-flash)