Artificial Analysis Cyber Index: Results as Published

The Artificial Analysis Cyber Index measures defensive work on real software: finding vulnerabilities, reproducing and checking them, and patching them without breaking existing behavior. It does not ask a model to build a working exploit. The index is an equal average of three scores, CWE-Bench-AA pass@1, DeepsecBench-AA F2 (the median of three runs), and CyberGym-E2E-AA pass@1, each placed on a 0 to 100 scale. A task declined on safety grounds scores zero, and safety blocks are also reported as their own average across the three tests.

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.