Google DeepMind's new benchmark tests agents on blueprints, schematics and polished decks

Google DeepMind launched the Multimodal Knowledge Agent benchmark (MKA) on November 10th, a 200-task test built with Artificial Analysis to grade how AI agents handle visually dense professional files such as blueprints, schematics and polished decks. Google says its Gemini 4 Carbon model ranks second on quality while completing tasks at six times lower cost on average than the top-scoring model, a claim the company frames as making price per completed task a metric to compare alongside peak score. The benchmark was announced by Google DeepMind senior product manager Rohan Doshi and research director Ben Caine.

Google DeepMind's new benchmark tests agents on blueprints, schematics and polished decks Built with Artificial Analysis, MKA uses 200 expert-designed tasks to grade visual knowledge work. Google says Gemini 4 Carbon ranks second on quality at a lower cost. By RuntimeWire Staff https://runtimewire.com/author/runtimewire-staff · Published · Updated Primary source: Google Gemini official https://blog.google/innovation-and-ai/models-and-research/gemini-models/multimodal-knowledge-agent-benchmark/ Why it matters MKA evaluates agents on visual files and finished deliverables that define professional work. For model buyers, Google's second-place quality ranking paired with a claimed sixfold cost advantage makes price per completed task a number to compare alongside peak score. Google DeepMind senior product manager Rohan Doshi and research director Ben Caine launched the Multimodal Knowledge Agent benchmark on November 10th, a 200-task test built with Artificial Analysis to grade how AI agents handle visually dense professional files. Google https://blog.google/innovation-and-ai/models-and-research/gemini-models/multimodal-knowledge-agent-benchmark/?ref=runtimewire says Gemini 4 Carbon https://runtimewire.com/models/google/gemini-4-carbon ranks second on quality in the test while completing tasks at six times lower cost on average than the top-scoring model.… When you purchase through links in our articles, we may earn a small commission https://runtimewire.com/affiliate-monetization-standards . This doesn’t affect our editorial independence.