Google DeepMind releases MKA benchmark testing agents on 10 professional visual work domains
_akhaliq · x · 2026-11-11
Google DeepMind released the Multimodal Knowledge Agent (MKA) benchmark on Hugging Face, evaluating autonomous agents on complex multi-step visual knowledge work across 10 professional domains.
- Tasks span GIS/surveying, architecture, electrical & RF engineering, mechanical engineering, and more, each mapped to a real job role
- Unlike OCR-solvable benchmarks, every task has a multi-file workspace with load-bearing visual artifacts: CAD blueprints, electrical/RF schematics, geospatial rasters, biochemical gel plates, multi-axis charts
- Requires native multimodal perception, cross-document synthesis, and tool orchestration (code execution, file inspection)
- Deliverables are standalone outputs: .xlsx, .pptx, .docx, .json, or structured technical reports; Apache 2.0 licensed
Related event: DeepMind and Artificial Analysis Launch MKA Benchmark for Multimodal Agents(6 posts)→
More from Research
- New paper shows benign interaction memories can lead agents to unsafe actions later — Amit_Goldenb · 2026-11-11
- Sliding Window Attention Explained with Cartoon Huangs: Old Tokens Slide Out of View — cfregly · 2026-11-11
- Goodfire releases deep neuron interpretability study with autonomous Neuron Scientist agent — teortaxesTex · 2026-11-11
- Baseten details deep research agent training: breadth-rewarding RL plus SFT for report writing — baseten · 2026-11-11
- onóna: twin embeddings with tied unembeddings cut validation loss across scales — Cheng Luo · 2026-11-11
- Caveats: small models only, and 50k+ poison examples can work without clean data — OwainEvans_UK · 2026-11-11