Unverified50% confidenceFactExact time
CVE-Bench is based on publicly disclosed CVE entries and tests a model's ability to identify, classify, and analyze the exploitability of known vulnerabilities.
1
Sources
50%
Confidence
Long-term
Relevance
10/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedDependency scanning tools partially rely on known vulnerability databases (such as CVE and OSV) for version comparison, and partially identify potentially malicious code by analyzing the dynamic behavioral characteristics of packages; the latter approach has a detection advantage over newly emerging counterfeit packages not yet in vulnerability databases.69% similarUnverifiedSAST discovers potential vulnerabilities by analyzing source code, while DAST discovers security issues through runtime probing; these methods are built around deterministic assumptions.68% similarUnverifiedSWE-bench相比HumanEval、MBPP等传统基准更贴近实际开发场景，考验模型对大型代码库的理解、跨文件依赖追踪和缺陷定位能力64% similarUnverified注意力机制驱动的语义理解能够识别传统正则匹配和静态分析工具无法发现的新型漏洞利用路径，在零日漏洞发现场景中具备优势62% similarUnverifiedThe agent runs project tests through the IDE, retrieves failure messages with line numbers, and traces issues to specific call sites in the source code.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/996931API
curl https://kongchang.com/api/v1/knowledge/claims/996931MCP
get_claim(id=996931)