Google DeepMind launches MKA, a benchmark for visually complex knowledge work
Google DeepMind and Artificial Analysis released MKA, a benchmark of 200 tasks that scores agents on visually complex professional work.
Google DeepMind, in partnership with Artificial Analysis, launched the Multimodal Knowledge Agent (MKA) benchmark, built to evaluate how agents reason across visually complex files to solve real-world knowledge work tasks. MKA was built with domain experts across 10 verticals and 83 occupations where day-to-day work is inherently visual.
Those specialists, from architects and biologists to underwriters and data scientists, created MKA’s 200-task evaluation set to assist them in their jobs. Each task reflects a multi-step project that takes a seasoned professional three to eight hours to complete.
Scoring and the first results
Each output is graded against an expert-written rubric of roughly 70 yes/no checks, weighted 75% on factual accuracy and 25% on visual design. When domain experts graded model outputs themselves, they agreed with the autorater 96% of the time. Among the models tested, Gemini 4 Carbon achieves #2 in quality and completes tasks at 6x lower cost on average compared to the top scoring model.
What to do
Artificial Analysis released a separate set of 10 public MKA tasks on Hugging Face, so anyone can inspect the format and test their own agents. Gemini 4 Carbon is available today via the Gemini API (application programming interface) in Google AI Studio.
Sources
Related reading
3 postsDiscussion
Spotted an error? Send corrections to [email protected] .