Cogent Security described VR-1, a post-trained cyber reasoning model built to investigate unfamiliar enterprise environments from a limited foothold, combine weaknesses across cloud, identity, application runtime, CI/CD, SaaS and internal documentation, and execute verified attack chains. On the black-box configuration of Cogent's own IntrusionBench, VR-1 reportedly achieved more than a 2x lift in pass@3 over the strongest evaluated frontier baseline within a two-hour wall-clock or 250-turn budget, with the gap narrowing in grey-box and disappearing largely in white-box settings. Cogent characterizes VR-1 as "Mythos-class" only in the sense of moving from finding weaknesses to executing material attack paths, not as parity with Anthropic's Mythos on binary exploitation or zero-day discovery. Both VR-1 and IntrusionBench are at early preview; results come from a small task set, with task counts and confidence intervals promised in a later technical report.