Benchmarking 13 AI models on rediscovering known CVEs
aikido.dev | blog | #ai-security | #vulnerability-research | #benchmark | #automation | #code-review | #cve | #llm | #vulnerability-detection
Summary
Aikido benchmarked 13 AI models on 26 known vulnerabilities: GPT-5.6 led at 88.5%, cheaper variants trailed by a finding or two, pooled pass@3 runs beat one flagship pass, and GLM-5.2 hit 16/26.
- Published
- Collected
--- id: 4609 title: "Benchmarking 13 AI models on rediscovering known CVEs" source_domain: aikido.dev source_url: "https://www.aikido.dev/blog/benchmarking-ai-models-known-cves" published_at: "2026-07-16 00:00" tags: ["ai-security", "vulnerability-research", "benchmark", "automation", "code-review", "cve", "llm", "vulnerability-detection"] phying_url: "https://phying.com/item/4609" lang: en --- # Benchmarking 13 AI models on rediscovering known CVEs ## Summary Aikido benchmarked 13 AI models on 26 known vulnerabilities: GPT-5.6 led at 88.5%, cheaper variants trailed by a finding or two, pooled pass@3 runs beat one flagship pass, and GLM-5.2 hit 16/26. ## Intelligence Metadata - **Sources:** aikido.dev - **Published:** 2026-07-16 00:00 - **Original:** https://www.aikido.dev/blog/benchmarking-ai-models-known-cves - **Phying News:** https://phying.com/item/4609