200 MKA entries kept private... scoring weight is actually 75% factual accuracy and 25% design

Google DeepMind and Artificial Analysis released MKA, a new benchmark of 200 tasks spanning 10 industry sectors and 83 job roles, designed to evaluate multimodal knowledge agents that turn visual inputs such as drawings and video into professional work products. Scoring weights factual accuracy at 75% and visual design at 25%, with each task modeled on multi-step work a skilled practitioner would take three to eight hours to complete. The benchmark matters because it shifts agent evaluation toward end-to-end document creation from visual materials rather than isolated perception or text tasks. What to watch next is which models are ranked on MKA and how scores on its 200 tasks track real-world adoption.