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
CWE-bench v2. Every model the catalog carries a published CWE-bench value for, ranked by that value.
Defensive vulnerability patching on 250 audit-and-patch tasks spanning 185 weakness types, 100 public and 150 held out. 20 models in the ModelChoice catalog have a published value.
Each point is a model with a published CWE-bench value and complete token prices, on the 75% input / 25% output blend. Frontier models are spaced evenly along the cost axis.
20 models with a published CWE-bench 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 | Anthropic | 75.0 | |
| 2 | 70.7 | ||
| 3 | 68.1 |
What the benchmark measures and how its value is published, from the source’s own record.
CWE-bench (CWE-bench v2) is an agentic benchmark published by CWE-bench v2 leaderboard and methodology. Defensive vulnerability patching on 250 audit-and-patch tasks spanning 185 weakness types, 100 public and 150 held out. Its evaluation set is described by the source as audit and patch C/C++, Go, Java, TypeScript/JavaScript, Python, and Rust repositories; 100 public and 150 held-out tasks. The source rates its difficulty as frontier agent evaluation.
Each model's result is published by CWE-bench v2 leaderboard and methodology as pass@1. ModelChoice prints that value to 1 decimal place, exactly as published, and never converts or rescales it. This page ranks the 20 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 | 67.3 |
| 5 | Anthropic | 67.1 |
| 6 | OpenAI | 66.8 |
| 7 | Meta | 66.1 |
| 8 | OpenAI | 66.0 |
| 9 | 65.0 |
| 10 | OpenAI | 64.2 |
| 11 | Tencent | 64.1 |
| 12 | Anthropic | 61.4 |
| 13 | OpenAI | 60.0 |
| 14 | DeepSeek | 57.6 |
| 15 | xAI | 54.7 |
| 16 | Tencent | 47.3 |
| 17 | Mistral | 45.6 |
| 18 | Alibaba | 45.4 |
| 19 | Moonshot AI | 45.3 |
| 20 | Z.AI | 42.0 |
Evidence key: Observed
Rows are ordered by the value CWE-bench v2 leaderboard and methodology 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.