Google and Artificial Analysis launch Multimodal Knowledge Agent benchmark MKA

On November 10, Google announced the launch of MKA (Multimodal Knowledge Agent), a benchmark developed jointly with the AI evaluation organization Artificial Analysis to assess how well AI agents handle visually complex documents and real-world knowledge work tasks. Google said its newly released Gemini 4 Carbon ranked second among participating models on the benchmark, while its average cost per completed task was only about one-sixth of the cost of the model ranked first. Rohan Doshi, Senior Product Manager at Google DeepMind, and Ben Caine, Lead Engineer, explained that many existing knowledge-work benchmarks can be completed using text alone, whereas real jobs often involve large amounts of visual information. Architects cross-check building blueprints, engineers analyze engineering schematics, and professionals consolidate information from different sources into slides, reports, and spreadsheets, they noted, adding that MKA is intended to measure AI agent performance in exactly these complex scenarios. To build the benchmark, Google worked with domain experts from 83 professions across 10 industries, producing an evaluation set of 200 tasks that spans business, architecture, education, engineering, finance, healthcare, law and government, logistics and maps, science, and software and data. As of November 11, the benchmark had been released by Google together with Artificial Analysis, with the evaluation set size and Gemini 4 Carbon's ranking and cost figures publicly disclosed; no further updates have been reported in the available material.