The best model per security task — measured, not assumed.

Every ranking below is a real benchmark run. Route each security sub-task to the model that empirically wins it — not the one you assume is best.