Google DeepMind and Artificial Analysis introduce Multimodal Knowledge Agent benchmark for visual professional workflows

What happened.Google DeepMind and Artificial Analysis launched the Multimodal Knowledge Agent benchmark to evaluate how autonomous AI agents execute visual professional projects across 83 occupations.
Why it matters.Traditional benchmarks rely on plain text or optical character recognition, hiding how badly autonomous agents struggle when reading dense visual schematics. Establishing rigorous rubrics for architectural blueprints and scientific charts gives engineering teams reliable data before deploying autonomous agents into visual design pipelines.
The technical specifications:
- Project scope: Assembles 200 challenging multi-step tasks requiring three to eight hours of human expert labor across architecture, biology, and finance.
- Visual artifacts: Requires models to inspect complex non-text inputs including computer-aided design blueprints, electrical schematics, and multi-axis charts.
- Standardized harness: Evaluates frontier models inside the open-source Stirrup execution sandbox equipped with code execution, web fetch, and image inspection tools.
- Dual scoring: Grades generated deliverables using an autorater aligned with human experts that weights factual accuracy at 75 percent and visual layout at 25 percent.
- Frontier divide: Highlights wide efficiency splits among top models, where Gemini 4 Carbon matches frontier satisfaction rates while consuming far fewer tokens.
STORYLINE PROGRESSION & CONTEXT
This report represents a verified evolution of prior newsroom intelligence: