Together with @ArtificialAnlys, we built MKA (Multimodal Knowledge Agent): a benchmark for evaluating agents' ability to handle visually complex tasks.
Together with @ArtificialAnlys, we built MKA (Multimodal Knowledge Agent): a benchmark for evaluating agents' ability to handle visually complex tasks.
Everyday professional work often uses more than just text—from cross-referencing blueprints to reading scientific charts. Existing evaluations do not cover these.
Gemini 4 Carbon ranks 2nd, with an average cost per task only one-sixth that of the top-ranked model. →