Curated by StartupHub.ai
Corma Security's DefenseBench.
Rate
Defensive Task Performance vs End-to-End Task Execution Speed
DefenseBench V5 by Corma Security
LowerDefensive Task PerformanceHigher
Claude Fable 5.2 (max): End-to-End Task Execution Speed: 15, Defensive Task Performance: 42.
Claude Opus 5.5 (max): End-to-End Task Execution Speed: 31, Defensive Task Performance: 36.
Gemini 4 Carbon (high): End-to-End Task Execution Speed: 28, Defensive Task Performance: 30.
GPT-6.1 Sol (max): End-to-End Task Execution Speed: 56, Defensive Task Performance: 31.
GPT-6 Astra (max): End-to-End Task Execution Speed: 56, Defensive Task Performance: 27.
SlowerEnd-to-End Task Execution SpeedFaster
End-to-End Task Execution Speed read from the chart you shared. Defensive Task Performance read from the chart you shared.

