Loading page…
Preferred languages
More languages supported by Google Translate
Open the selector to load the current language catalog.
Translations are provided by Google Translate.
Loading page…
Preferred languages
More languages supported by Google Translate
Open the selector to load the current language catalog.
Translations are provided by Google Translate.
Agentic benchmark
Artificial Analysis CWE-Bench-AA. Every model the catalog carries a published CWE-Bench-AA value for, ranked by that value.
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. 21 models in the ModelScale catalog have a published value.
Each point is a model with a published CWE-Bench-AA value and complete token prices, on the 75% input / 25% output blend. Frontier models are spaced evenly along the cost axis.
21 models with a published CWE-Bench-AA value, ordered by that value, highest first. Models the source has not scored on this benchmark are not listed — they are unmeasured, not last.
| Rank | Model | Provider | Pass@1 |
|---|---|---|---|
| 1 | DeepSeek | 68.3 | |
| 1 | xAI | 68.3 | |
| 3 | OpenAI | 65.0 |
What the benchmark measures and how its value is published, from the source’s own record.
CWE-Bench-AA (Artificial Analysis CWE-Bench-AA) is an agentic benchmark published by CWE-Bench-AA Benchmark Leaderboard. 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. Its evaluation set is described by the source as real-world vulnerability patching tasks. The source rates its difficulty as automated vulnerability remediation.
Each model's result is published by CWE-Bench-AA Benchmark Leaderboard as pass@1. ModelScale prints that value to 1 decimal place, exactly as published, and never converts or rescales it. This page ranks the 21 models that carry a published value, highest value first; a model the source never scored is left out rather than ranked at zero. The source does not state whether a higher value is the better result, so neither does this page.
| 4 | OpenAI | 64.2 |
| 5 | OpenAI | 63.3 |
| 5 | Xiaomi | 63.3 |
| 7 | OpenAI | 61.7 |
| 8 | Moonshot AI | 60.8 |
| 9 | Anthropic | 58.3 |
| 9 | Mistral | 58.3 |
| 11 | OpenAI | 56.7 |
| 12 | Z.AI | 55.8 |
| 12 | Z.AI | 55.8 |
| 14 | Moonshot AI | 53.3 |
| 15 | 52.5 |
| 15 | StepFun | 52.5 |
| 17 | Anthropic | 50.8 |
| 18 | RABeam 1.0 501B-A23B | Reflection AI | 49.2 |
| 19 | MiniMax | 41.7 |
| 20 | NVIDIA | 27.5 |
| 21 | Alibaba | 26.7 |
Evidence key: Observed
Rows are ordered by the value CWE-Bench-AA Benchmark Leaderboard published, highest first. The source does not state whether a higher value is the better result, so this page does not either: for a benchmark that measures a rate of failure, read the table from the bottom.