BlogLLM Benchmark Scores as Published, Side by Side
CVE-Bench: Results as Published
Every result the catalog holds for CVE-Bench: 3 models from 1 developer, read from 1 page. Each number is as printed by the company named beside it; nothing is ranked or rescored. Models are listed by developer, newest first.
Area CybersecurityMaintainer Yuxuan Zhu, Daniel Kang and co-authorsIntroduced 2025Data version 2026103001, as of 2026-10-30
CVE-Bench measures whether AI agents can succeed against 40 real, serious web application vulnerabilities taken from the US national vulnerability database. The targets are content management systems and AI applications, and each attempt is checked automatically inside a sandbox. The paper's headline finding is that the best agent framework succeeded on up to 13 percent of the vulnerabilities.
CVE-Bench reward
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| Grok 4.7xAI, released 2026-09-21 | 37.7% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
| 36.6% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: xhigh | |
| Grok 4.6xAI, released 2026-08-12 | 39.8% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
| Grok 4.5xAI, released 2026-07-16 | 35.2% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
Every number here is read from the same catalog file as the benchmark table, data version 2026103001, and is the number the table shows for that model and test. Where one page printed a model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.