PWNBench prices live agent security

PWNBench, Novee’s new live-agent penetration-testing benchmark, reports Claude Opus 5 at 51% recall for roughly $1,400 at k=3, versus Kimi K3 at 42% for $209. Novee’s benchmark release says the test measures live web applications and exposes tradeoffs across recall, precision, and spend; it does not establish a universal model ranking. Security teams should treat domain-specific evaluation spend as part of model selection, not a post-launch audit, as the agent security control gap makes clear.