EVERY SIGNAL, SORTED.

Google DeepMind, Artificial Analysis, Multimodal Knowledge Agent Benchmark, Gemini 4 Carbon, Knowledge Work, Visual Documents, Spreadsheets, Presentations, Reports, Professional Evaluation, Agentic AI, Human Review and Responsible AI

Google built a benchmark for visual office work. Keep the artifact in the grade

A new benchmark asks agents to inspect dense professional files and return real spreadsheets, presentations, reports and marked-up images. The useful shift is simple: grade the file people must actually use, not the confident sentence that says the work is done.