Moonshot AI's PerceptionBench Shows No Frontier Model Tops 60% on Pure Visual Perception

Moonshot AI's new PerceptionBench finds every frontier model scores under 60% on pure visual-perception tasks.