| Mobile Web

Google DeepMind releases benchmark for professional tasks reading blueprints and clinical images

Google DeepMind has released a new benchmark to assess whether AI agents can perform real professional work, going beyond answering simple questions. The Multimodal Knowledge Agent benchmark, developed with Artificial Analysis, uses complex visual inputs such as architectural drawings, circuit designs and clinical images and requires agents to produce deliverables like reports and spreadsheets. It spans 200 tasks across 83 job roles, scoring factual accuracy and visual quality.