# Artificial Analysis CWE-Bench-AA (CWE-Bench-AA)

> Artificial Analysis' implementation of Collinear AI's CWE-bench, testing whether agents can patch real-world software vulnerabilities. One of the three components of the AA Cyber Index.

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

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

## About CWE-Bench-AA

- Year: 2026
- Tasks: Real-world vulnerability patching tasks
- Format: Pass@1
- Difficulty: Automated vulnerability remediation
- Paper: [CWE-Bench-AA Benchmark Leaderboard](https://artificialanalysis.ai/evaluations/cwe-bench-aa)

Separate from the provider-reported CWE-Bench rows. Independently run by Artificial Analysis under its published harness. Display-only; not yet admitted as independent-run evidence.

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

## Leaderboard (19 models)

| Rank | Model | Creator | Score |
|------|-------|---------|-------|
| 1 | [DeepSeek V4.1 Flash 1017](/models/deepseek-v4-1-flash) | DeepSeek | 68.3% |
| 2 | [Grok 4.7](/models/grok-4-7) | xAI | 68.3% |
| 3 | [GPT-6 Sol](/models/gpt-6-sol) | OpenAI | 64.2% |
| 4 | [GPT-6 Astra](/models/gpt-6-astra) | OpenAI | 63.3% |
| 5 | [MiMo-V2.6-Pro](/models/mimo-v2-6-pro) | Xiaomi | 63.3% |
| 6 | [Kimi K3.1 Pro](/models/kimi-3-1-pro) | Moonshot AI | 60.8% |
| 7 | [Claude Opus 5.5](/models/claude-opus-5-5) | Anthropic | 58.3% |
| 8 | [GPT-6 Luna](/models/gpt-6-luna) | OpenAI | 56.7% |
| 9 | [GLM-5.3](/models/glm-5-3) | Z.AI | 55.8% |
| 10 | [GLM-5.3-Flash](/models/glm-5-3-flash) | Z.AI | 55.8% |
| 11 | [Kimi K3](/models/kimi-k3) | Moonshot AI | 53.3% |
| 12 | [Gemini 3.8 Flash](/models/gemini-3-8-flash) | Google | 52.5% |
| 13 | [Step 5 Preview](/models/step-5-preview) | StepFun | 52.5% |
| 14 | [Claude Fable 5.1](/models/claude-fable-5-1) | Anthropic | 50.8% |
| 15 | [Mistral Large 4](/models/mistral-large-4) | Mistral | 50.8% |
| 16 | [Beam 1.0 501B-A23B](/models/reflection-beam) | Reflection AI | 49.2% |
| 17 | [MiniMax M3](/models/minimax-m3) | MiniMax | 41.7% |
| 18 | [Nemotron 3 Ultra](/models/nemotron-3-ultra) | NVIDIA | 27.5% |
| 19 | [Qwen3.8-27B](/models/qwen3-8-27b) | Alibaba | 26.7% |

## FAQ

### What does CWE-Bench-AA measure?

Artificial Analysis' implementation of Collinear AI's CWE-bench, testing whether agents can patch real-world software vulnerabilities. One of the three components of the AA Cyber Index.

### Which model scores highest on CWE-Bench-AA?

DeepSeek V4.1 Flash 1017 by DeepSeek currently leads with a score of 68.3% on CWE-Bench-AA.

### How many models are evaluated on CWE-Bench-AA?

19 AI models have published results on CWE-Bench-AA in the BenchLM catalog.

### Does CWE-Bench-AA affect BenchLM's overall score?

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

## Compare Top Models on CWE-Bench-AA

- [DeepSeek V4.1 Flash 1017 vs Grok 4.7](/compare/deepseek-v4-1-flash-vs-grok-4-7)
- [Grok 4.7 vs GPT-6 Sol](/compare/gpt-6-sol-vs-grok-4-7)
- [GPT-6 Sol vs GPT-6 Astra](/compare/gpt-6-astra-vs-gpt-6-sol)
- [GPT-6 Astra vs MiMo-V2.6-Pro](/compare/gpt-6-astra-vs-mimo-v2-6-pro)